fix(caveman-stats): correct Opus output price ($75→$25) for 4.5+ era

[jmmarquez2025]

Fixes #465.

Problem

src/hooks/caveman-stats.js priceForModel() prefix-matches every claude-opus-4* id to $75/M output — the legacy Claude 3 Opus rate. Anthropic's Opus rate card has been $5/M in, $25/M out since Opus 4.5 (4.8 shipped on the same card, 2026-05-28). So /caveman-stats overstates the savings/cost figure ~3× for anyone running a current Opus model — increasingly the default.

claude-haiku-4 was also stale ($4 vs Haiku 4.5's $5/M).

Fix

Split the table into legacy vs current Opus, most-specific prefixes first (the loop returns the first match), so retired 4.0/4.1 (incl. dated ids like claude-opus-4-20250514) keep $75 while 4.5–4.8 resolve to $25:

['claude-opus-4-0',    75.00],
['claude-opus-4-1',    75.00],
['claude-opus-4-2025', 75.00],   // dated 4.0 ids
['claude-opus-4',      25.00],   // 4.5–4.8
...
['claude-haiku-4',      5.00],   // Haiku 4.5

This keeps the existing dated-id test (claude-opus-4-20250101 → 75) valid instead of regressing it — preferred over a flat constant flip, which is the alternative I floated in #465.

Tests

tests/test_caveman_stats.js updated and green — 29/29. Added asserts pinning both eras (claude-opus-4-8 → 25, claude-opus-4-1-20250805 → 75).

Note: npm test (the tests/installer/* suite) shows 4 failing opencode … tests, but they fail identically on a clean main checkout — pre-existing and unrelated to this change.

Source: https://platform.claude.com/docs/en/about-claude/pricing

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.