Add Ollama provider for local model support

[tusharadi25]

Summary

• Add an opt-in Ollama provider so DataStoria's AI features can run against locally hosted models (Llama, Qwen, Mistral, DeepSeek, etc.).
• The provider is off by default and enabled by setting OLLAMA_BASE_URL. Optional OLLAMA_API_KEY supports deployments where Ollama sits behind an auth proxy.
• Locally available models are discovered automatically via Ollama's native GET /api/tags endpoint and surfaced as system-backed models; chat/completions go through an OpenAI-compatible client (/v1).
• Base URL accepts either the bare host (http://localhost:11434) or the OpenAI-compatible base (.../v1); discovery failures degrade gracefully (empty list, never throws) so the rest of the model catalog keeps loading.
• GitHub Copilot and Ollama both expose models dynamically, so Ollama is exempted from static-catalog verification in createModel.
• Docs/UX: provider link to the Ollama library in model settings, .env.example entry, and README/docs mentions.

Changes

• src/lib/ai/llm/ollama-models.ts — host normalization, config detection, /api/tags discovery, model normalization.
• src/lib/ai/llm/llm-provider-factory.ts — register Ollama provider (OpenAI-compatible), system key gated on OLLAMA_BASE_URL, skip catalog verification.
• src/lib/ai/llm/provider-ids.ts — add PROVIDER_OLLAMA.
• src/app/api/ai/models/available/route.ts — append discovered Ollama models to system models (fetched in parallel with GitHub models).
• src/components/settings/models/models-edit.tsx — provider link for Ollama.
• public/provider-logos/ollama.svg, .env.example, README.md, docs/index.md.

Test plan

• npm run test — unit tests for Ollama discovery/normalization (ollama-models.test.ts), provider registration and dynamic model resolution (llm-provider-factory.test.ts), and the available-models route (route.test.ts).
• npm run build / npm run typecheck — verify build and types.
• Manual: with OLLAMA_BASE_URL=http://localhost:11434 and a pulled model, confirm the model appears in settings and chat works; with the env unset, confirm Ollama does not appear.

Note: npm run build was not run locally in this environment (system Node is v10, and the Docker build hits a pre-existing @openrouter/ai-sdk-provider vs ai@6 ERESOLVE conflict on master, unrelated to this change which does not touch package.json).

[vercel]

@tusharadi25 is attempting to deploy a commit to the Frank Chen's projects Team on Vercel.

A member of the Team first needs to authorize it.