feat(ai-review): estimate costs across the model pricing catalog

[absorbb]

AI review cost estimates were populated only for Terra, leaving the new Astra default without a cost. Add a 47-model dictionary from the official OpenAI standard text pricing tables, verified on 2026-10-16, covering the current review models and published legacy/specialized text models.

The estimator prices each request using cumulative usage deltas, ignores repeated token-count events, separates uncached input/cache reads/cache writes/output, and applies the published long-context premium above 272,000 input tokens per request. Reasoning output is already included in output tokens. Unknown exact model identifiers, unpriced usage categories, or inconsistent counters produce an unavailable reason rather than an invented cost.

The workflow summary and analysis artifact include the pricing source, verification date, standard-rate basis, cache-write usage, and any unavailable reason. Estimates exclude tools, regional/service-tier uplifts, and the optional native benchmark. Pricing entries do not imply Codex model compatibility.

Source: https://developers.openai.com/api/docs/pricing

Validation: all 42 tests pass, including model-specific rates, cache writes, context boundaries, duplicate events, and unavailable cases; actionlint and git diff --check pass. A token-only replay of 256 real Codex usage events also completed with an estimate. Fresh local medium-tier review checked all 47 rates and found no issues.

[jitsu-code-review]

Review of e67a0c6f7d2cc226e7108afc55c4152806cf2a58

Reviewed the full diff, workflow integration, and official pricing. No high-confidence defects found. All 42 tests pass.