feat(ai): add Ollama provider integration with parallel LLM calls

[aGallea]

Summary

• Replace free-text provider input with a closed dropdown (OpenAI, Google, Anthropic, Ollama)
• When Ollama is selected, fetch available models from local instance via POST /api/ai/ollama/models and display as selectable dropdown
• Skip API key requirement for Ollama across frontend (Settings, ClusterLegend, ClusterDetailDrawer) and backend (litellm calls)
• Normalize base URL by stripping /v1 suffix for Ollama models since litellm's ollama/ provider uses the native API, not the OpenAI-compatible endpoint
• Parallelize cluster naming LLM calls using asyncio.gather + thread pool executor for significantly faster naming (N clusters in ~1 call duration instead of N x call duration)

Changed Files

Backend:

• ai_naming.py — Add _normalize_base_url(), conditionally pass api_key to litellm
• server/models.py — Add OllamaModelsRequest, OllamaModel, OllamaModelsResponse
• server/routes/ai.py — Add Ollama models endpoint, parallelize name_clusters and name_sub_clusters

Frontend:

• api/ai.ts — Add AI_PROVIDERS constant, AiProvider type, fetchOllamaModels()
• pages/SettingsPage.tsx — Provider dropdown, Ollama model dropdown, conditional API key
• components/plot/ClusterLegend.tsx — Allow naming without API key for Ollama
• components/plot/ClusterDetailDrawer.tsx — Allow sub-cluster naming without API key for Ollama
• types/index.ts — Add OllamaModel, OllamaModelsResponse types

Test Plan

• All 265 tests pass
• mypy + ruff clean
• Pre-commit hooks pass
• Manual E2E: Select Ollama -> models load -> Name with AI works without API key

[github-actions]

Tests Report 📄

Tests Succees ✅

JUnit Details

Total Tests	Failures	Errors	Skipped	Time ⏳
281	0	0	0	71.83s

Coverage Details (100% >= 90%) ✅

Diff Cover Details

File	Covered Lines		Missing Lines
embedding_cluster/ai_naming.py	19/19	100%
embedding_cluster/server/models.py	15/15	100%
embedding_cluster/server/routes/ai.py	86/86	100%
Total	120/120	100%