fix(mcp): preserve cache invalidation during refresh

[fengting124]

This pull request fixes an MCP tools-cache race where an in-flight
list_tools() refresh could clear a newer invalidation or leave a retained
stale schema authoritative for call_tool() validation.

Each refresh captures the current cache generation and only publishes its
result when that generation remains current. While the cache is dirty, local
required-parameter validation is skipped until a later refresh establishes a
clean authoritative snapshot.

A deterministic concurrency regression covers the stale refresh result,
retained cache snapshot, dirty-cache call-through, subsequent refresh, and
restored clean-cache validation.

Validation:

• targeted concurrency and validation regression
• full tests/mcp/test_caching.py
• related filtering, pagination, retry, and validation tests
• make format
• make lint
• make typecheck
• make tests
• git diff --check

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 36e85b1dff

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".

[fengting124]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

Reviewed commit: cbbcd49b53

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: eb1c136e94

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]

Let a surviving refresh publish after a newer attempt fails

When two cache-miss list_tools() calls overlap in the same generation, the second call immediately changes _tools_refresh_sequence; if that newer call is then cancelled or raises while the first call succeeds, this condition prevents the successful survivor from publishing. The server therefore retains a stale or absent _tools_list, remains dirty, performs another remote fetch on the next listing, and skips local required-parameter validation in the interim. Track the newest successful refresh rather than letting a failed attempt permanently disqualify surviving work, and cover the A pending -> B starts -> B fails -> A succeeds ordering.

AGENTS.md reference: AGENTS.md:L104-L104

Useful? React with 👍 / 👎.

[seratch]

Please remove _tools_refresh_sequence and use the captured invalidation generation to decide whether a completed refresh can publish. With cache_tools_list=False, a newer failed or cancelled refresh currently prevents an older successful refresh from updating the cache. That can leave an obsolete schema marked clean, causing call_tool() to reject arguments that match the successfully returned tool definition.

Please add a regression covering a previously cached schema requiring q, an updated schema without that requirement, and the ordering A pending -> B starts -> B fails -> A succeeds. Assert that the subsequent valid tool call reaches the server.