fix(chat): support DeepSeek reasoning models by preserving reasoning_content

[Frankli9986]

Closes #53

Problem

Users configuring DeepSeek (e.g. deepseek-reasoner / V4 thinking models) hit the API error:

The reasoning_content in the thinking mode must be passed back to the API.

This happens on the second turn of a chat session. The current code routes DeepSeek through the generic @ai-sdk/openai provider, which does not recognize DeepSeek's reasoning_content. As a result:

1. The first response's reasoning_content is dropped.
2. On the next request convertToModelMessages() produces an assistant message without reasoning_content.
3. DeepSeek API rejects the request.

Changes

Dependencies

• Add @ai-sdk/deepseek for first-class DeepSeek API support.

Settings / provider selection

• Add deepseek as a first-class AI provider in settings-store.ts and settings-dialog.tsx.
• Default config: https://api.deepseek.com + deepseek-chat.

Model list endpoint

• src/app/api/ai/models/route.ts: add DeepSeek /models lookup.

Provider factory

• src/lib/ai/provider.ts: route deepseek provider to createDeepSeek().

Chat streaming

• src/app/api/ai/chat/route.ts:
  • Enable sendReasoning: true so reasoning tokens flow to the client as UI message parts.
  • Persist reasoning in chatMessages.metadata for historical message round-trips.

Message conversion

• src/lib/ai/utils.ts: restore { type: 'reasoning' } parts from DB metadata.

Tests

• src/lib/ai/provider.test.ts: verify DeepSeek model creation.
• src/lib/ai/utils.test.ts: verify reasoning part round-trip.

Verification

1. Settings → AI Provider → select DeepSeek
2. Enter API key and model (e.g. deepseek-reasoner or deepseek-chat)
3. Start a chat, ask AI to fill resume
4. Continue the conversation — no reasoning_content error on subsequent turns

Scope note

This is intentionally focused on the chat route because issue #53 is a chat/session error. The other generateText call sites (tools.ts for analyzeJdMatch and translateResume) do not maintain multi-turn context, so they are not affected by the reasoning_content requirement.

[vercel]

@Frankli9986 is attempting to deploy a commit to the twwch's projects Team on Vercel.

A member of the Team first needs to authorize it.