Reduce LLM max tokens from 128K to 8K for reasoning

[rysweet]

Change DEFAULT_LLM_MAX_TOKENS from 128000 to 8192 for reasoning (JSON response is always under 500 tokens). Add a separate TRANSCRIPT_MAX_TOKENS=128000 for transcript analysis if needed later.

Files to modify:

• src/amplihack/fleet/_constants.py
• src/amplihack/fleet/_backends.py

Requirements:

1. Change DEFAULT_LLM_MAX_TOKENS = 8192 in _constants.py
2. Add TRANSCRIPT_MAX_TOKENS = 128000 in _constants.py
3. Update _backends.py to use DEFAULT_LLM_MAX_TOKENS for reasoning calls
4. Update _backends.py to use TRANSCRIPT_MAX_TOKENS for transcript analysis calls
5. Add docstrings explaining the distinction
6. All 918+ fleet tests must pass
7. Commit changes on this branch