test(community-v3): cover OpenAPI and other runtime-state tools

[naji247]

Summary

• Closes the coverage gap that let the context-injection regression fixed in fix(community-v3): inject tool context without deep-copying the tool #38 ship green: the community suite only ever exercised plain @server.tool function tools (which deep-copy cleanly), so no test caught that FastMCP v3 tools holding live runtime state fail to deep-copy.
• Adds end-to-end tests that drive the real middleware dispatch (tools/list + tools/call) against OpenAPI-generated, proxy, and mounted tools — the three tool types that hold non-deepcopyable runtime state (an httpx.AsyncClient, a client factory, a live sub-server reference respectively).
• Runs entirely in-process via httpx.MockTransport — no network, no live backend.
• Test-only change. No production code is modified.

What's covered

tests/test_utils/community_openapi_server.py (new factories):

• create_community_openapi_server(record_requests=…) — multi-endpoint OpenAPI spec over a mock transport; tools are real OpenAPITools.
• create_community_proxy_server() (ProxyTool) and create_community_mounted_server() (FastMCPProviderTool).

tests/community/test_community_v3_openapi.py (end-to-end):

• context injected into every tool on tools/list, with the custom description and marked required
• no "Error copying tool" logged
• server's cached tools are not mutated across repeated lists
• tools/call strips context, captures it as user_intent, and never leaks it to the downstream HTTP request
• HTTP-500 tool calls are captured as errors and surface to the client
• the tools/list event response serializes OpenAPITool subclasses
• get_more_tools coexists with generated tools and is excluded from injection
• deepcopy precondition guard

tests/community/test_community_v3_runtime_state_tools.py:

• parametrized guard over [openapi, proxy, mounted] asserting context injection with zero copy failures.

All modules are FastMCP v3-only and skip cleanly when FastMCP is absent (the test-without-fastmcp job) and on the v2 compatibility matrix.

Test plan

• uv run pytest tests/community/test_community_v3_openapi.py tests/community/test_community_v3_runtime_state_tools.py -v — 12 passed on FastMCP 3.0.0b1
• Regression check: temporarily restoring copy.deepcopy(tool) in the middleware fails 7 of the new tests (including the OpenAPI and mounted injection guards), confirming they catch the exact regression from fix(community-v3): inject tool context without deep-copying the tool #38
• uv run pytest tests/community/ tests/test_context_parameters.py -q — 97 passed (was 85), no regressions
• uv run --with 'fastmcp==3.4.4' pytest tests/community/ tests/test_context_parameters.py -q — 97 passed on the upper matrix bound
• Confirmed the new modules import cleanly and skip when FastMCP is not installed