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

frontend/src/lib/components/llm/ProviderConfig.svelte

@@ -40,6 +40,8 @@ import { SvelteSet } from "svelte/reactivity";
     is_active: true,
     is_multimodal: false,
     max_images: null,
+    context_window: null,
+    max_output_tokens: null,
   });
 
   let testStatus = $state({ type: "", message: "" });
@@ -84,6 +86,8 @@ import { SvelteSet } from "svelte/reactivity";
       is_active: true,
       is_multimodal: false,
       max_images: null,
+      context_window: null,
+      max_output_tokens: null,
     };
     editingProvider = null;
     testStatus = { type: "", message: "" };
@@ -109,6 +113,8 @@ import { SvelteSet } from "svelte/reactivity";
       is_active: Boolean(provider?.is_active),
       is_multimodal: Boolean(provider?.is_multimodal),
       max_images: provider?.max_images ?? null,
+      context_window: provider?.context_window ?? null,
+      max_output_tokens: provider?.max_output_tokens ?? null,
     };
     testStatus = { type: "", message: "" };
     availableModels = [];
@@ -271,6 +277,18 @@ import { SvelteSet } from "svelte/reactivity";
       delete submitData.api_key;
     }
 
+    // Normalize token limit fields: bind:value on <input type="number"> returns string or "";
+    // Pydantic expects int | None, so convert empty/string to number or null.
+    for (const field of ["context_window", "max_output_tokens"]) {
+      const val = submitData[field];
+      if (val === "" || val === null || val === undefined) {
+        submitData[field] = null;
+      } else if (typeof val === "string") {
+        const num = Number(val);
+        submitData[field] = Number.isNaN(num) ? null : num;
+      }
+    }
+
     try {
       await requestApi(endpoint, method, submitData);
       showForm = false;
@@ -570,6 +588,51 @@ import { SvelteSet } from "svelte/reactivity";
               </div>
             {/if}
           </div>
+
+          <!-- #region token_limits_advanced -->
+          <!-- @BRIEF Collapsible "Advanced: Token Limits" section. context_window and max_output_tokens
+                     per-provider override for token budget estimation. NULL = use PROVIDER_DEFAULTS. -->
+          <details class="border-t pt-3 mt-2">
+            <summary class="text-sm font-medium text-text-muted cursor-pointer hover:text-text select-none outline-none">
+              {$t.llm?.advanced_token_limits || "Advanced: Token Limits"}
+            </summary>
+            <div class="mt-2 space-y-2">
+              <div>
+                <label for="provider-context-window" class="block text-sm font-medium text-text">
+                  {$t.llm?.context_window_label || "Context Window (tokens)"}
+                </label>
+                <input
+                  id="provider-context-window"
+                  type="number"
+                  min="1000"
+                  max="256000"
+                  placeholder="{$t.llm?.auto_detect || "Auto"}"
+                  bind:value={formData.context_window}
+                  class="mt-1 block w-full border rounded-md p-2"
+                />
+                <p class="mt-0.5 text-xs text-text-subtle">
+                  {$t.llm?.context_window_hint || "Leave blank for auto-detection from model name."}
+                </p>
+              </div>
+              <div>
+                <label for="provider-max-output-tokens" class="block text-sm font-medium text-text">
+                  {$t.llm?.max_output_label || "Max Output Tokens"}
+                </label>
+                <input
+                  id="provider-max-output-tokens"
+                  type="number"
+                  min="256"
+                  placeholder="{$t.llm?.auto_detect || "Auto"}"
+                  bind:value={formData.max_output_tokens}
+                  class="mt-1 block w-full border rounded-md p-2"
+                />
+                <p class="mt-0.5 text-xs text-text-subtle">
+                  {$t.llm?.max_output_hint || "Must be less than context window. Leave blank for auto."}
+                </p>
+              </div>
+            </div>
+          </details>
+          <!-- #endregion token_limits_advanced -->
         </div>
 
         {#if testStatus.message}