Add GPT-5.6 pricing

[Finesssee]

Summary

• Ports upstream aa40f1eb4 GPT-5.6 Sol, Terra, and Luna standard rates plus >272,000 whole-request long-context rates into the canonical CostUsagePricing map.
• Normalizes gpt-5.6, openai/gpt-5.6, supported Codex suffixes, and dated aliases to gpt-5.6-sol through one final alias application.
• Preserves per-token-count JSONL records through CodexUsageRecord, so multiple short requests do not become one incorrectly long-context-priced daily aggregate.

Upstream mapping

Upstream behavior	Win-CodexBar implementation/tests
aa40f1eb4: Sol/Terra/Luna standard input, cache-read, output rates	CODEX_PRICING; test_gpt56_standard_pricing
aa40f1eb4: whole-request long-context rates over 272k	Optional CodexLongContextRates on the same canonical model entry; long-context and threshold tests
aa40f1eb4: gpt-5.6 aliases resolve to Sol	Alias tests include combined OpenAI/Codex/date form
Per-request long-context accounting	JSONL records retain token-count deltas; two 200k Sol requests total $2.00 instead of one false 400k long-context request
501e95150: stabilization	No numeric GPT-5.6 rate change. Models.dev pricing-key and Pi cache invalidation are non-applicable: this port has neither subsystem.

Structural follow-up

• Retired unused packed day/model state from CodexParseResult and CodexParserState; all four parser tests assert records.
• Consolidated both parser record appends in record_usage.
• Long-context rates now live on their canonical model entry. Models with no optional tier explicitly retain standard rates above the threshold.

Unsupported dimensions

• Cache-write: current Codex JSONL records contain only input, cached-read, and output.
• Priority: upstream derives it from separate SQLite trace metadata (service_tier and turn IDs); this port has no such scanner or record fields.
• Pi repricing: no Pi scanner/cache exists under rust/src.
• No zero-cost fallback was added; usage remains visible when pricing is unavailable.

Validation

• RED/GREEN: initial GPT-5.6 rate and alias tests failed before implementation; the final-alias path failed during the structural refactor until suffix normalization retained gpt-5.6 for final aliasing.
• Focused: cargo test --manifest-path rust/Cargo.toml gpt56 --lib (5); cargo test --manifest-path rust/Cargo.toml 'core::jsonl_scanner::tests' --lib (8).
• Full shared: fmt check; cargo test --manifest-path rust/Cargo.toml (536 library + 1 binary); clippy -D warnings.
• Full Tauri: fmt check; cargo test --manifest-path apps/desktop-tauri/src-tauri/Cargo.toml (303); clippy -D warnings.
• git diff --check passed. Frontend validation is not applicable because no frontend or bridge file changed.