Enforce one provider request per R2 slot

[iotaverbum-core]

Closes #63

Defect corrected

The historical pilot path performs a model-availability GET before the Responses API POST. R2 authorizes one provider request per slot, so that inherited preflight would exceed scope.

Correction

• adds openai_gpt56_r2_single_request.py as the only canonical R2 execution entrypoint;
• replaces the network model preflight with an offline exact-request binding;
• leaves the single Responses API POST as the only reachable provider HTTP request per slot;
• preserves the exact requested alias in the sealed request;
• preserves the provider-returned model label in capture metadata;
• binds both the canonical wrapper and underlying control-plane module into the benchmark lock;
• restores all patched references after every invocation;
• adds offline regression tests proving the historical preflight function is not called.

Authority reset

No R2 provider request or block authorization was made before this correction. Any prior human authorization is not used. After merge and pull, block 1 requires a fresh explicit authorization naming the corrected entrypoint and merged commit.

This PR itself makes no provider request and grants no execution, judgment, ratification, publication, promotion, ranking, release, or later-block authority.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 68adabb82e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Disable the legacy R2 CLI path

In a campaign-machine environment where an operator follows the existing docs/openai-gpt56-sol-memory-boundary-r2-harness.md commands, this new canonical-command statement is not enforced: openai_gpt56_r2.py is still documented and executable, and its execute-next --execute path still delegates to openai_live_pilot.main, which performs the model GET before the Responses POST. That leaves a reachable two-provider-request R2 slot despite this correction, so the legacy CLI/doc path needs to be retired or made to fail closed.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Bind authorizations to the wrapper and commit

The promised fresh authority reset is not machine-enforced: authorize-block through the wrapper delegates to the same core implementation, and the block-authorization record still contains no corrected-entrypoint or merged-commit field, only the plan/slots/policy/operator data. In a campaign machine with a block authorization created before this correction or by the legacy CLI, execute-next through the wrapper accepts the same canonical path and cannot distinguish stale authority, so the authorization schema/read check should bind and verify the corrected entrypoint and commit.

Useful? React with 👍 / 👎.