pricing: gpt-5.6-codex and gpt-5.6-codex-max missing from the bundled snapshot (price at $0)

[iamtoruk]

Found during #1075 verification: gpt-5.6-codex and gpt-5.6-codex-max have no entry in src/data/litellm-snapshot.json, so their calls price at $0.00 — 285 sessions / 5,446 calls in one real corpus. Adjacent to #1047/#1056 (codex activity-id pricing) but distinct: these are real model ids emitted by Codex, not activity aliases. Fix: regenerate/extend the snapshot (or add explicit fallback rows) with sourced rates.

[iamtoruk]

Correction after empirical reconciliation on the cited corpus: this issue conflated two things during the #1075 verification pass. The "285 sessions / 5,446 calls at $0" are codex-auto-review calls (now 5,717/279 with corpus growth) — a distinct raw activity id, already tracked in #1047 and fixed by open PR #1056 (prices it via gpt-5.5; verified +$780.24 on this corpus). The literal gpt-5.6-codex/gpt-5.6-codex-max ids appear only in 6 aborted stub sessions with zero token_count events, and main's longest-prefix pricing fallback already resolves them to the bare gpt-5.6 rates anyway. PR #1090 still adds explicit snapshot rows for both ids (evidenced by the 4-generation pattern of codex-suffixed ids billing identically to their bare siblings, and robust against future snapshot regens) — correct hardening, but it moves $0 on this corpus. The dollars this issue was actually about belong to #1056.