fix(proxy-responses): adapt explicit prompt cache controls

[Komzpa]

Summary

Prevent public explicit prompt-cache controls from causing 400 responses on
the Codex subscription upstream. The subscription egress now omits unsupported
controls while preserving prompt text, order, and prompt_cache_key; successful
fallbacks report their effective semantics in a response header.

Head: ecc281960393c1be719dcfb07236c0d19744f3ef

Type of change

• fix: - bug fix (no behavior change beyond the bug)
• feat: - new user-facing feature or capability
• refactor: - internal refactor (no behavior change, no API change)
• docs: - documentation only
• chore: / ci: / build: - tooling, CI, packaging
• test: - test-only change
• Breaking change

Linked issue: None; this is a focused Responses compatibility regression.

OpenSpec

• This PR includes / updates an OpenSpec change
• Not applicable - bug fix that matches the existing spec
• Not applicable - docs / CI / chore only
• This PR touches a codex-faithful path and preserves upstream-equivalent behavior

Change directory: openspec/changes/adapt-subscription-prompt-cache-controls/

The intentional divergence is limited to Codex subscription egress, whose
upstream rejects these public API controls. OpenAI-compatible model-source
egress keeps the public request unchanged.

Changes

• Strip prompt_cache_options and supported content-block
  prompt_cache_breakpoint fields only during subscription serialization.
• Preserve explicit controls for OpenAI-compatible model sources.
• Return X-Codex-LB-Prompt-Cache-Mode: subscription-implicit after a
  successful subscription fallback so clients can observe the semantic
  downgrade.
• Cover both routes with unit and /v1/responses integration regressions.

Simplicity

• Works with zero configuration.
• Adds no required setup step or setting.
• New settings: none.
• Adds no README section, .env.example entry, or dashboard navigation.

Test plan

uv run pytest tests/unit/test_strict_schema_validation.py tests/integration/test_openai_compat_features.py
# 89 passed

make lint
# proxy architecture checks passed; Ruff check and format check passed across 858 files

uv run ty check
# All checks passed

uv run openspec validate adapt-subscription-prompt-cache-controls --strict
# Change is valid

uv run pre-commit run local-ci --hook-stage manual --all-files
# passed

Output

A successful subscription request containing an explicit cache breakpoint now
returns:

X-Codex-LB-Prompt-Cache-Mode: subscription-implicit

The corresponding subscription wire payload retains prompt_cache_key and
content but omits the unsupported explicit controls. The model-source negative
control retains all three fields and does not return the fallback header.

Checklist

• Title is in Conventional Commits format.
• No related public issue exists.
• Added tests covering the change.
• Ran the relevant local CI gate.
• Strict OpenSpec validation passes.
• Simplicity gates reviewed.
• CHANGELOG.md is unchanged.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[Komzpa]

@codex review

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.