fix: migrate from deprecated chat/completions to Responses API

[blackjr21]

Summary

xAI deprecated the /v1/chat/completions endpoint which now returns 410 "Live search is deprecated". This PR migrates grok-cli to use the /v1/responses endpoint.

Changes

src/grok/client.ts

1. chat() method - Now uses /v1/responses with native fetch instead of OpenAI SDK's chat.completions.create()

2. Tool format conversion - Flattened from nested structure:

   // Before (OpenAI format)
   { type: "function", function: { name, description, parameters } }
   
   // After (xAI Responses API format)  
   { type: "function", name, description, parameters }

3. Response parsing - Added convertResponseToGrokResponse() to extract text from xAI's response format:

   output[].content[].text  // where type === "output_text"

4. Streaming - Rewrote chatStream() to parse SSE format from Responses API:

   event: response.output_text.delta
   data: {"delta": "Hello"}
   

   Handles event types:

   • response.output_text.delta - Main content streaming
   • response.content_part.delta - Alternative content format
   • response.output_item.added/done - Tool calls
   • response.completed/done - Response finished

Testing

Tested and working:

• ✅ Basic chat (headless mode: grok -p "hello")
• ✅ Interactive mode (grok)
• ✅ Tool calling (view_file, bash, create_file, str_replace_editor)
• ✅ Streaming responses

Related Issues

Fixes #134

Notes

• This is a more complete fix than PR Replace deprecated Live Search API with Agent Tools API #132 which added agentQuery() but didn't switch the main conversation loop
• Maintains backward compatibility with existing code that uses the GrokResponse format

[homanp]

Fixed in #145