fix(claude): drop invalid tool history before upstream

[yinkev]

Fixes #4112.

Problem

Anthropic rejects stale or malformed Claude message history when:

• an assistant tool_use has no matching result in the immediately following user message;
• a user tool_result references no tool call in the previous assistant message; or
• a tool_result appears after text or image content instead of in the leading result block group.

The existing Claude-upstream sanitizer removed incompatible reasoning signatures but forwarded those structurally invalid tool blocks unchanged, producing HTTP 400 before model execution.

Change

Enable tool-history repair only in SanitizeClaudeMessagesForClaudeUpstream:

• pair assistant tool calls with leading results in the immediately following user message;
• preserve valid parallel tool calls and allow their results in any order;
• remove unmatched calls, stale results, blank IDs, wrong-role tool blocks, and results appearing after non-result content;
• remove messages left empty by that repair;
• retain the existing signature sanitizer behavior and preserve an unchanged payload without copying it when history is already valid.

Other provider compatibility paths keep their existing behavior because the new option is disabled by default.

Verification

• red/green sanitizer regressions for orphan calls/results, late results, partial parallel matches, and valid parallel history;
• outbound ClaudeExecutor regression confirming invalid blocks are absent from the Anthropic request body;
• existing Claude signature logging fixture updated to use valid paired tool history;
• go test ./internal/signature ./internal/runtime/executor -count=1;
• focused go test -race for the sanitizer and Claude executor paths;
• go vet ./internal/signature ./internal/runtime/executor;
• required go build -o test-output ./cmd/server && rm test-output;
• git diff --check.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 82ef4be747

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]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a93e8f3476

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]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a0081b93e6

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]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8b1a108174

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".