fix(spawn): pass max effort to Codex

[JaredHuynhning]

Summary

• Pass the shared max effort through Codex launches as model_reasoning_effort="max".
• Accept Codex max in the duplicated standing crew-dispatch validator and cover direct and configured dispatch paths with behavior tests.
• Keep ultra unsupported and rejected before metadata or launch construction.
• Correct the Codex adapter evidence and align the adjacent Pi verification comment with Pi and pi-signed 0.82.0 verified on 2026-07-27.

Measured behavior

On 2026-08-01, codex-cli 0.146.0 reported effort support per model. Max was verified for gpt-5.6-sol and gpt-5.6-terra.

A live trivial launch using gpt-5.5 with model_reasoning_effort="max" exited 1 with HTTP 400, error code unsupported_value, parameter reasoning.effort, and an explicit supported-value list of none, low, medium, high, and xhigh. No assistant response or work was produced.

Firstmate deliberately remains model-agnostic instead of maintaining a changing model-capability map. An unsupported model/effort pairing fails loudly and actionably rather than silently clamping or performing work at a different effort.

Validation

The no-mistakes run completed fresh review and fix re-review with no remaining findings, targeted spawn and bootstrap behavior tests, documentation validation, and the pinned repository lint gate.

CI note

This pull request comes from a fork. GitHub requires a maintainer to approve workflow execution, so checks will initially appear as not run.