feat(codex): add GPT-6.1 Sol and update client to 0.159.2

[fenix007]

The stable fork does not list GPT-6.1 Sol, and its fallback Codex identity and optional CLI image still use 0.156.1. This adds the model and aligns the client pins at 0.159.2.

• Add OpenAI/Codex catalog entries with separate context limits, native Codex routing, Max/Ultra handling, low Codex default, Responses tool routing and Fast eligibility.
• Accept Responses endpoints when importing discovered Codex models.
• Add Standard short-context price estimates ($2 input / $0.10 cached / $10 output per million tokens), a 2x purchased-credit Fast multiplier and Sol-specific protection against counting reasoning-inclusive output twice. Included-subscription allowance consumption and long-context billing are outside this change.
• Preserve caller-provided client versions, connection/request reasoning preferences and the newly merged account/model eligibility behavior.

Adapted from upstream #15171 at 11143fed867b4104c6964851de9b7d0b99d7f6a4 and the discovery validation fix in #15158. Thanks to @xiaoyaner0201 and @HouMinXi. Adapted to the existing frozen fork modules rather than importing the newer upstream refactors.

Validation:

• Focused Codex, catalog, discovery, identity, account-model access, schema, SSE/tool and pricing regressions passed: 146 tests across 18 test files. SSE/tool traffic is mocked; no live account requests were made.
• Core TypeScript, changed-file ESLint, provider consistency, test discovery and documentation synchronization passed.
• The file-size gate is already failing on the exact base 7e20ea6273505fa5f674f531b4fbb08d562b7236: accountFallback.ts is 2035 lines against 1953, and account-fallback-service.test.ts is 2111 against 1982. Both files and their baseline are unchanged here.
• An expanded TypeScript check including the new test has the same 27 inherited production diagnostics as the exact base, zero additional diagnostics, and no diagnostics in the new test.
• Full unit/coverage/build runs were not performed during the concurrent PR sync. No quality thresholds or baselines are changed.

Prepared in a separate worktree and branch. This PR does not merge into stable, publish an image or deploy production.