Add Ollama support as a local model provider

[tusharadi25]

Summary

Add Ollama as a supported model provider so users can run DataStoria's AI features (natural-language query generation, optimization, visualization, diagnostics) against local, self-hosted models instead of only cloud APIs.

Today the README lists OpenAI, Anthropic Claude, Google Gemini, GitHub Copilot, OpenRouter, Groq, Cerebras, and Nebius under Multiple Model Providers — but there is no option for a fully local/offline backend.

Motivation

• Privacy / data residency — Many ClickHouse operators work with sensitive data and cannot send schema, queries, or results to a third-party cloud API. Ollama keeps everything on-prem.
• Cost — No per-token billing for high-volume, multi-turn agent conversations.
• Offline / air-gapped environments — Enables DataStoria in environments without outbound internet access.
• Experimentation — Easy to try open models (Llama 3.x, Qwen, Mistral, DeepSeek, etc.) for SQL generation/optimization.

Proposed Solution

• Add an Ollama provider in the model-provider configuration UI/settings.
• Configurable base URL (default http://localhost:11434) to support remote/containerized Ollama servers.
• Model selection/discovery via Ollama's GET /api/tags (list locally available models).
• Use Ollama's OpenAI-compatible endpoint (/v1/chat/completions) where possible to minimize new client code, falling back to the native /api/chat API if needed.
• No API key required (optional bearer token for proxied/secured deployments).
• Ensure streaming responses and the existing Agent SKILLs flow work with the Ollama backend.

Additional Context

• Ollama exposes an OpenAI-compatible API, which should make integration relatively low-effort if the existing OpenAI client can be pointed at a custom base URL.
• References: Ollama API docs, Ollama OpenAI compatibility.

Happy to help test against a local Ollama instance.