Make AI token budget limits user-configurable

[heysamtexas]

Summary

Currently AI token budget limits are hard-coded in the library API (50-1000 tokens). This should be configurable by users while maintaining safe defaults.

Current Implementation

# Hard-coded in src/git_summary/library.py
@field_validator("ai_token_budget")
@classmethod  
def validate_token_budget(cls, v: int) -> int:
    if v < 50:
        raise ValueError("ai_token_budget must be at least 50 for basic analysis")
    if v > 1000:
        raise ValueError("ai_token_budget cannot exceed 1000 to avoid rate limits")
    return v

Proposed Solution

1. Add AI Configuration to Config Class

• Extend ~/.git-summary/config.json to include AI settings
• Add methods: get_ai_config(), set_ai_config(), get_token_budget_limits()
• Structure: {"ai": {"min_token_budget": 50, "max_token_budget": 1000, "default_token_budget": 200}}

2. Update Library API Validation

• Replace hard-coded limits in AnalysisConfig.validate_token_budget()
• Load limits from user config with fallback to safe defaults
• Add validation that user limits are reasonable (e.g., max can't exceed 10000)

3. Add CLI Commands for Configuration

git-summary config ai --min-budget 25 --max-budget 2000
git-summary config ai --reset  # restore defaults
git-summary config ai --show   # display current limits

4. Enhanced Validation & Warnings

• Warn users when they set high limits about cost implications
• Add --force flag to override safety warnings
• Include budget guidance in dry-run output

Benefits

• Flexibility: Power users can set higher limits when needed
• Safety: Defaults remain conservative, changes require explicit action
• Transparency: Users see and control their own risk tolerance
• Enterprise-friendly: Organizations can set company-wide policies

Acceptance Criteria

• User can configure min/max token budget limits via CLI
• Configuration persists in ~/.git-summary/config.json
• Library API respects user-configured limits
• Safe defaults maintained (50-1000 tokens)
• Warnings for potentially expensive configurations
• Dry-run shows current budget limits and guidance

Priority

Low-medium - current hard-coded limits work well for 95% of users. Good candidate for community contribution after core library API is complete.