feat(ai): add VertexAI provider for Claude models

[rodaddy]

Summary

This PR adds support for Google Cloud Vertex AI as a provider to access Claude models using Application Default Credentials (ADC). This allows users to route their Fabric requests through Google Cloud Platform instead of directly to Anthropic, enabling billing through GCP.

Features

• Support for Claude models (Sonnet 4.5, Opus 4.5, Haiku 4.5, 3.7 Sonnet Latest, 3.5 Haiku Latest) via Vertex AI
• Uses Google Application Default Credentials for authentication (no API keys required)
• Configurable project ID and region (defaults to 'global' for cost optimization)
• Full support for streaming and non-streaming requests
• Implements complete ai.Vendor interface

Configuration

Environment variables:

• VERTEXAI_PROJECT_ID: GCP project ID (required)
• VERTEXAI_REGION: Vertex AI region (optional, defaults to 'global')

Example .env configuration:

DEFAULT_VENDOR=VertexAI
DEFAULT_MODEL=claude-sonnet-4-5
VERTEXAI_PROJECT_ID=your-gcp-project
VERTEXAI_REGION=global

Testing

Tested with:

• Claude Sonnet 4.5 model
• Global endpoint (us-east5 also tested)
• Both streaming and non-streaming requests
• YouTube transcript processing

Related Issues

Closes #1570

Checklist

• Follows contributing guidelines (small PR, conventional commits)
• Implements complete ai.Vendor interface
• Registered in plugin_registry.go
• Tested with real-world usage
• Changelog will be generated after PR number is assigned

[ksylvan]

@rodaddy Can you retest with the fixes/refactor I made to the PR?

[rodaddy]

@ksylvan Reviewed the refactoring - nice work on the toMessages extraction!

Verified:

• ✅ Code compiles cleanly
• ✅ Message conversion logic properly handles:
  • System message prepending to first user message
  • User/assistant role alternation enforcement
  • Empty message filtering
  • Edge cases (assistant-first, system-only scenarios)
• ✅ Response handling iterates all content blocks (vs just first)

working on testing your code now

[rodaddy]

@ksylvan Retested with the refactored code - works perfectly! ✅

Test Results:

Non-streaming request:

echo "Summarize quantum computing in one sentence" | fabric --model claude-sonnet-4-5 --vendor VertexAI

✅ Response received successfully

Streaming request:

echo "Write a haiku about coffee" | fabric --model claude-sonnet-4-5 --vendor VertexAI --stream

✅ Streaming works perfectly

Code Quality:
The toMessages refactor is a major improvement:

• ✅ Proper system message handling (prepended to first user message)
• ✅ User/assistant role alternation enforcement
• ✅ Empty message filtering
• ✅ Better response content block iteration
• ✅ Improved error handling

The refactored implementation is solid and ready to merge from my perspective.

[ksylvan]

Approved @rodaddy - Thanks for the contribution!

[rodaddy]

@ksylvan, will I get on the contributor list for this?

[ksylvan]

Yes, @rodaddy - Happy New Year!