perf(translate): fix slow translation startup — CJK estimation, output budget, provider token config

Root cause: batch sizing underestimated CJK token density (1.5→1.0 chars/token)
and ignored output budget as primary constraint, causing cascading finish_reason=length.

Changes:
- _token_budget.py: CJK_RATIO 1.5→1.0, OTHER_RATIO 2.2→1.8, safety factors 0.75/0.70
- _token_budget.py: new _compute_max_rows_by_output() — output budget is PRIMARY constraint
- _batch_sizer.py: resolve_provider_config() with DB-level context_window/max_output_tokens
- _batch_sizer.py: INPUT_SAFETY_FACTOR applied, max_rows_by_output used as row cap
- _llm_http.py: log actual usage.prompt_tokens/.completion_tokens from provider
- _llm_call.py: retry only missing rows after finish_reason=length (save partial result)
- models/llm.py + schema: provider-level context_window / max_output_tokens (nullable)
- services/llm_provider.py: get_provider_token_config() helper
- Alembic migration: add columns to llm_providers
- Svelte ProviderConfig: collapsible Advanced: Token Limits section
- 12 new tests (token budget, batch sizer, provider config)
- All 492 tests pass

backend/src/services/__tests__/test_llm_provider.py

@@ -380,3 +380,121 @@ def test_llm_provider_config_multimodal_explicit():
         is_multimodal=True,
     )
     assert config.is_multimodal is True
+
+
+# endregion test_llm_provider_config_multimodal_explicit
+
+# region test_llm_provider_config_context_window_default [TYPE Function]
+# @RELATION  BINDS_TO -> [LLMProviderConfig]
+# @PURPOSE: Verify LLMProviderConfig.context_window defaults to None.
+def test_llm_provider_config_context_window_default():
+    """Verify default context_window is None in schema."""
+    config = LLMProviderConfig(
+        provider_type=LLMProviderType.OPENAI,
+        name="Default Test",
+        base_url="https://api.openai.com/v1",
+        api_key="sk-test",
+        default_model="gpt-4",
+    )
+    assert config.context_window is None
+
+
+# endregion test_llm_provider_config_context_window_default
+
+# region test_llm_provider_config_context_window_explicit [TYPE Function]
+# @RELATION  BINDS_TO -> [LLMProviderConfig]
+# @PURPOSE: Verify LLMProviderConfig accepts explicit context_window value.
+def test_llm_provider_config_context_window_explicit():
+    """Verify setting context_window explicitly works."""
+    config = LLMProviderConfig(
+        provider_type=LLMProviderType.OPENAI,
+        name="Test",
+        base_url="https://api.openai.com/v1",
+        api_key="sk-test",
+        default_model="gpt-4",
+        context_window=128000,
+        max_output_tokens=16384,
+    )
+    assert config.context_window == 128000
+    assert config.max_output_tokens == 16384
+
+
+# endregion test_llm_provider_config_context_window_explicit
+
+# region test_get_provider_token_config_no_provider [TYPE Function]
+# @RELATION  BINDS_TO -> [LLMProviderService]
+# @PURPOSE: Verify get_provider_token_config returns all-None when provider not found.
+def test_get_provider_token_config_no_provider():
+    """When provider_id doesn't exist, returns all values as None."""
+    db = MagicMock(spec=Session)
+    db.query.return_value.filter.return_value.first.return_value = None
+
+    service = LLMProviderService(db)
+    result = service.get_provider_token_config("nonexistent-id")
+    assert result == {"model": None, "context_window": None, "max_output_tokens": None}
+
+
+# endregion test_get_provider_token_config_no_provider
+
+# region test_get_provider_token_config_with_values [TYPE Function]
+# @RELATION  BINDS_TO -> [LLMProviderService]
+# @PURPOSE: Verify get_provider_token_config returns provider token limits from DB.
+def test_get_provider_token_config_with_values():
+    """Provider with context_window and max_output_tokens returns them."""
+    db = MagicMock(spec=Session)
+    mock_provider = MagicMock(spec=LLMProvider)
+    mock_provider.default_model = "gpt-4o-mini"
+    mock_provider.context_window = 128000
+    mock_provider.max_output_tokens = 16384
+    db.query.return_value.filter.return_value.first.return_value = mock_provider
+
+    service = LLMProviderService(db)
+    result = service.get_provider_token_config("provider-1")
+    assert result["model"] == "gpt-4o-mini"
+    assert result["context_window"] == 128000
+    assert result["max_output_tokens"] == 16384
+
+
+# endregion test_get_provider_token_config_with_values
+
+# region test_get_provider_token_config_null_limits [TYPE Function]
+# @RELATION  BINDS_TO -> [LLMProviderService]
+# @PURPOSE: Verify get_provider_token_config returns None for null DB fields.
+def test_get_provider_token_config_null_limits():
+    """Provider with NULL token limits returns None values (signal to use defaults)."""
+    db = MagicMock(spec=Session)
+    mock_provider = MagicMock(spec=LLMProvider)
+    mock_provider.default_model = "qwen-flash"
+    mock_provider.context_window = None
+    mock_provider.max_output_tokens = None
+    db.query.return_value.filter.return_value.first.return_value = mock_provider
+
+    service = LLMProviderService(db)
+    result = service.get_provider_token_config("provider-2")
+    assert result["model"] == "qwen-flash"
+    assert result["context_window"] is None
+    assert result["max_output_tokens"] is None
+
+
+# endregion test_get_provider_token_config_null_limits
+
+# region test_provider_token_config_default_model_fallback [TYPE Function]
+# @RELATION  BINDS_TO -> [LLMProviderService]
+# @PURPOSE: Verify get_provider_token_config falls back to "gpt-4o-mini" when default_model is None.
+def test_provider_token_config_default_model_fallback():
+    """Provider without explicit default_model uses 'gpt-4o-mini' fallback."""
+    db = MagicMock(spec=Session)
+    mock_provider = MagicMock(spec=LLMProvider)
+    mock_provider.default_model = None
+    mock_provider.context_window = None
+    mock_provider.max_output_tokens = None
+    db.query.return_value.filter.return_value.first.return_value = mock_provider
+
+    service = LLMProviderService(db)
+    result = service.get_provider_token_config("provider-3")
+    assert result["model"] == "gpt-4o-mini"
+    assert result["context_window"] is None
+    assert result["max_output_tokens"] is None
+
+
+# endregion test_provider_token_config_default_model_fallback

backend/src/services/llm_provider.py

@@ -115,6 +115,8 @@ class LLMProviderService:
                 is_active=config.is_active,
                 is_multimodal=config.is_multimodal,
                 max_images=config.max_images,
+                context_window=config.context_window,
+                max_output_tokens=config.max_output_tokens,
             )
             self.db.add(db_provider)
             self.db.commit()
@@ -148,6 +150,8 @@ class LLMProviderService:
             db_provider.is_active = config.is_active
             db_provider.is_multimodal = config.is_multimodal
             db_provider.max_images = config.max_images
+            db_provider.context_window = config.context_window
+            db_provider.max_output_tokens = config.max_output_tokens
 
             self.db.commit()
             self.db.refresh(db_provider)
@@ -235,6 +239,24 @@ class LLMProviderService:
 
     # endregion get_decrypted_api_key
 
+    # region get_provider_token_config [TYPE Function]
+    # @PURPOSE: Returns provider token limits for batch sizing.
+    # @PRE  provider_id must be valid.
+    # @POST  Returns dict with model name, context_window, max_output_tokens.
+    #        Values from DB take priority; None means "use PROVIDER_DEFAULTS fallback".
+    # @RATIONALE Centralised helper — both _batch_proc.py and _batch_sizer.py need
+    #            the same resolution logic. Avoids duplicating DB queries and defaults.
+    def get_provider_token_config(self, provider_id: str) -> dict:
+        provider = self.get_provider(provider_id)
+        if not provider:
+            return {"model": None, "context_window": None, "max_output_tokens": None}
+        return {
+            "model": provider.default_model or "gpt-4o-mini",
+            "context_window": provider.context_window,
+            "max_output_tokens": provider.max_output_tokens,
+        }
+    # endregion get_provider_token_config
+
 
 # #endregion LLMProviderService