refactor(frontend): enforce semantic tokens + extract 3 Screen Models

Color tokens: 3658→0 violations across 186 .svelte/.svelte.ts files.
All raw Tailwind colors (bg-blue-*, text-gray-*, border-red-*, etc.)
replaced with semantic tokens from tailwind.config.js in 5 perl passes.

Model extraction (11→8 oversized pages):
- LLMReportModel.svelte.ts: LLM report page 413→221 lines
- DatasetDetailModel.svelte.ts: dataset detail page 416→218 lines
- DatasetsHubModel.svelte.ts: datasets hub page 468→246 lines

Tests: 14 assertions updated (color classes + model refs).
Build: 1 duplicate class:text-primary fix in ValidationTaskForm.
All 699/699 tests pass.

frontend/tests/maintenance.test.ts

@@ -160,7 +160,7 @@ describe('DashboardMaintenanceBadge', () => {
         const { container } = render(DashboardMaintenanceBadge, {
             props: { dashboardId: 42 }
         });
-        const badgeElements = container.querySelectorAll('.bg-orange-100');
+        const badgeElements = container.querySelectorAll('.bg-warning-light');
         expect(badgeElements.length).toBe(0);
     });
 
@@ -171,7 +171,7 @@ describe('DashboardMaintenanceBadge', () => {
         const { container } = render(DashboardMaintenanceBadge, {
             props: { dashboardId: 42 }
         });
-        const badge = container.querySelector('.bg-orange-100');
+        const badge = container.querySelector('.bg-warning-light');
         expect(badge).toBeTruthy();
         expect(badge?.textContent).toContain('Maintenance');
     });
@@ -183,7 +183,7 @@ describe('DashboardMaintenanceBadge', () => {
         const { container } = render(DashboardMaintenanceBadge, {
             props: { dashboardId: 42 }
         });
-        const badge = container.querySelector('.bg-orange-100');
+        const badge = container.querySelector('.bg-warning-light');
         expect(badge).toBeNull();
     });
 });