Add structured error propagation for API failures

[bledden]

Summary

Fixes #62 - Generic "Unable to generate final synthesis" error now shows actual error details.

Problem

When API calls fail, users see a generic message that hides the actual cause:

Error: Unable to generate final synthesis.

This makes it impossible to diagnose issues like invalid API keys, rate limits, or missing models.

Solution

backend/openrouter.py:

• Added ModelQueryError dataclass with error types
• Handle specific HTTP status codes:
  • 401: Auth error
  • 402: Payment required
  • 404: Model not found
  • 429: Rate limit exceeded
  • 5xx: Server error
  • Timeout handling

backend/council.py:

• Propagate errors through all 3 stages
• Aggregate errors in metadata
• Human-readable error summaries

Validation

# BEFORE (generic message):
User sees: "Error: Unable to generate final synthesis."

# AFTER (specific message):
User sees: "Error: Invalid API key. Please check your OPENROUTER_API_KEY."
# Plus structured error details in metadata.errors

Test plan

• Verify 401 errors show "Invalid API key" message
• Verify 429 errors show "Rate limit exceeded" message
• Verify 404 errors show which model was not found
• Verify errors are aggregated in metadata.errors

🤖 Generated with Claude Code