Document actual error response format from Artemis #12

Description

The error documentation should reflect the actual error response format returned by Artemis/LiteLLM, not just theoretical OpenAI-compatible errors.

Current State

src/content/docs/api/errors.md (if it exists) likely documents generic OpenAI error format.

Required Content

1. Actual Error Responses

   • Verify the exact JSON structure returned by Artemis
   • Document all error types actually used in production

2. Error Categories

   • Authentication Errors (401)

     • Invalid API key
     • Missing Authorization header
     • Expired key

   • Validation Errors (400)

     • Invalid model name
     • Missing required fields
     • Invalid parameter types

   • Rate Limit Errors (429)

     • Rate limit response format
     • Retry-After header

   • Server Errors (500)

     • vLLM unavailable (failover to Claude)
     • Internal errors

3. Error Handling Examples

   try:
       response = client.chat.completions.create(...)
   except AuthenticationError as e:
       # Handle invalid API key
   except RateLimitError as e:
       # Handle rate limit

4. Failover Behavior

   • Document that chat completions fail over to Claude Haiku when vLLM is down
   • Users should be aware of potential model changes during outages

Acceptance Criteria

• Error documentation matches actual Artemis response format
• All HTTP status codes documented (400, 401, 429, 500, 503)
• Real-world error examples from production
• Python error handling examples
• JavaScript error handling examples
• Documents failover behavior

Priority

MEDIUM - Important for production robustness.