feat(extension): upgrade AI SDK to v3 and improve Ollama integration

[lcomplete]

Summary

• Upgrade AI SDK packages from v1 to v3
• Replace ollama-ai-provider with OpenAI-compatible API for better Ollama support
• Add thinking mode persistence across sessions
• Fix streaming preview for new AI SDK chunk types

Test plan

• Test AI provider connection with OpenAI, Anthropic, Google, DeepSeek, Groq
• Test Ollama with local instance
• Verify thinking mode toggle persists after extension restart
• Test streaming preview for reasoning models

🤖 Generated with Claude Code

[augmentcode]

🤖 Augment PR Summary

Summary: Upgrades the browser extension’s AI integration to the newer AI SDK provider packages and improves Ollama support via an OpenAI-compatible endpoint.

Changes:

• Bump @ai-sdk/* provider packages to v3 and update provider model creation to use the new LanguageModel type.
• Replace ollama-ai-provider usage with createOpenAI pointed at an Ollama OpenAI-compatible /v1 base URL, plus helper URL normalization utilities.
• Add helper + tests for normalizing Ollama base URLs for both /api/tags and OpenAI-compatible chat endpoints.
• Persist “thinking mode” toggle state in chrome.storage.sync and wire it into both popup and preview UIs.
• Update streaming preview handling to account for new chunk shapes (adding chunk.text) and reasoning delta chunk types.
• Introduce a shared max-output-token constant used across streaming and OpenAI-compatible streaming paths.

Technical Notes: Streaming is processed via fullStream and preview state is derived from incremental chunk application; Ollama model discovery now uses a normalized base URL for /api/tags.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. 1 suggestion posted.

Comment augment review to trigger a new review at any time.

[augmentcode]

streamText().fullStream chunks are documented to use type: "text" (with text) for text deltas; applyStreamingPreviewChunk currently only treats "text-delta" as response text, so "text" chunks would be ignored and the preview may never show the actual answer.

Severity: high

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}