Add GPT-5.6 pricing

[bobergda]

Summary

• add API-equivalent pricing for GPT-5.6 Sol, Terra, and Luna
• apply GPT-5.6 long-context pricing above 272K input tokens: 2× input and cached-input rates, 1.5× output rate
• parse and price GPT-5.6 cache writes at 1.25× the applicable uncached-input rate
• add aliases for GPT-5.6 model names and regression coverage for all pricing paths

Why

Codex session logs already contain gpt-5.6-terra, but the dashboard treated it as unpriced. The initial fix omitted the model family’s long-context and cache-write billing rules, which underreported larger requests.

Validation

• PYTHONPATH=src python3 -m unittest tests.test_cli -v