fix: proper HTTP error handling across all LLM and VLM providers to prevent HTML leakage

[Wanbogang]

Problem

Fixes #1218

When api.openmind.org returns a 502 (or other HTTP errors), all LLM and VLM
providers were catching the error with a generic except Exception as e handler
and logging f"... error: {e}". This caused raw HTML responses to leak directly
into terminal output and log files.

This is especially critical for VLM providers which process frames continuously
at 10–30 fps — during an upstream outage, this could flood log files with HTML
and potentially fill disk, causing system crash.

Root Cause

All 10 affected files used a single generic exception handler that did not
distinguish between HTTP status errors, connection errors, and unexpected errors.

Fix

Replaced the generic handler with three specific handlers across all affected files:

except openai.APIStatusError as e:
    logging.error(f"... API error: HTTP {e.status_code} - {e.message}")
except openai.APIConnectionError as e:
    logging.error(f"... connection error: {e.message}")
except Exception as e:
    logging.error(f"... unexpected error: {type(e).__name__}: {e}")

Also added import openai to VLM providers that only had
from openai import AsyncOpenAI.

Files Changed

LLM Plugins:

• src/llm/plugins/openai_llm.py
• src/llm/plugins/openrouter.py
• src/llm/plugins/gemini_llm.py
• src/llm/plugins/deepseek_llm.py
• src/llm/plugins/xai_llm.py
• src/llm/plugins/near_ai_llm.py
• src/llm/plugins/qwen_llm.py

VLM Providers:

• src/providers/vlm_openai_provider.py
• src/providers/vlm_gemini_provider.py
• src/providers/vlm_openai_rtsp_provider.py

Notes

• Claim 2 (avatar fires while offline) and Claim 3 (infinite retry loop) from
  the original bug report were investigated and found to be invalid — cortex
  already guards with if output is None: return and there is no retry loop,
  only normal tick continuation.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!