[codex] Support DeepSeek V4 Pro, Grok 4.6, and tiered pricing

[LeenHawk]

What changed

• add official model metadata and request shaping support for DeepSeek V4 Pro/Flash
• add Grok 4.6 model metadata, reasoning support, and 500K context information
• preserve DeepSeek reasoning_effort and raise its documented output limit to 384K
• add generic multi-level prompt-token pricing tiers across billing, persistence, import/export, API, and console UI
• import OpenRouter pricing overrides, including 59 tiered models and multi-threshold Qwen rules
• enrich upstream model parsing with display names and explicit thinking capability metadata

Why

Grok 4.6 uses higher token prices after 200K prompt tokens, but this behavior is not xAI-specific. OpenAI, Google, Qwen, Seed, and other model families use different thresholds and sometimes multiple tiers. A generic tier representation avoids hard-coded model multipliers and keeps billing aligned with current provider catalogs.

Impact

Existing price rules continue to use their base rates. Rules may optionally provide pricing_tiers_json; the highest matching prompt-token threshold overrides only the rate categories it declares. Old SQLite/libSQL schemas are repaired automatically, and migration coverage verifies tier JSON can be written and read after upgrade.

Validation

• cargo check --all-targets --all-features
• cargo fmt --all -- --check
• database migration tests: 11 passed
• billing tier tests: 2 passed
• console Vitest suite: 22 files / 81 tests passed
• frontend typecheck, i18n parity, OpenRouter generator tests, and production build passed during implementation
• git diff --check