hermes - 💡(How to fix) Fix Make max_tokens configurable for aux vision calls

Problem

tools/vision_tools.py hardcodes max_tokens in two places for aux vision calls:

• Line 803: "max_tokens": 2000
• Line 1301: "max_tokens": 4000

Models like MiMo 2.5 support 130K output tokens, but vision descriptions are silently clipped to 2K/4K. These are the only aux paths where the cap is constraining — title gen (500), web extract (10K/20K), and context compressor (dynamic) are already adequate.

Proposed fix

Read from auxiliary.vision.max_tokens in config.yaml, falling back to current hardcoded values:

# config.yaml
auxiliary:
  vision:
    max_tokens: 8000   # optional, defaults to 2000/4000 per code path

# Line 803
- "max_tokens": 2000,
+ "max_tokens": _vision_cfg.get("max_tokens", 2000),

# Line 1301
- "max_tokens": 4000,
+ "max_tokens": _vision_cfg.get("max_tokens", 4000),

Why

Lets users get richer vision descriptions from capable models without editing upstream code that gets wiped on every update. Current defaults are preserved — just make them configurable.