feat: improve local LLM recommendations

[atomantic]

Summary

• Reframe local LLM recommendations around a general-purpose primary lane plus specialized use-case filters.
• Surface Qwen3.8 as the Best overall local pick across general, coding, reasoning, and vision recommendations.
• Preserve the scheduled catalog-refresh prompt upgrade path and add coverage.

Test plan

• cd server && npx vitest run lib/localLlmCatalog.test.js services/huggingFaceCatalog.test.js services/taskPromptDefaults.test.js services/taskSchedule.test.js
• cd client && npx vitest run src/components/settings/LocalLlmTab.test.jsx src/pages/LocalLlmPlayground.test.jsx
• cd client && npm run lint
• cd client && npm run build