fastapi_startkit: pin to 0.47.0 (FastAPI 0.139) to match baseline

[bedus-creation]

Summary

Aligns the committed fastapi_startkit benchmark with the fastapi baseline on the same FastAPI version (task the-benchmarker#670, decision c).

Previously python/fastapi_startkit/pyproject.toml used fastapi-startkit[fastapi]>=0.46,<0.47, which resolves startkit 0.46.x → FastAPI 0.124.4 / Starlette 0.50, while the fastapi baseline entry runs FastAPI 0.139 / Starlette 1.3.1 — an apples-to-oranges comparison (and a mismatch vs the published README numbers, which were measured on 0.139).

Change

• Pin to fastapi-startkit[fastapi]==0.47.0 → FastAPI 0.139.0 / Starlette 1.3.1, matching the baseline.
• Idiomatic include_router registration is kept unchanged. The flat-route candidate (PR fastapi_startkit: flat route registration (avoid include_router) — candidate #8) was investigated and deliberately not adopted: on 0.124 it's a structural no-op, and on 0.139 it bypasses startkit's representative router registration to dodge an intended FastAPI behavior (_IncludedRouter nested resolution). The ~10-14% include_router cost is a real, representative cost on 0.139 and is shown, not hidden.

Verification

• Docker image resolves fastapi 0.139.0, fastapi-startkit 0.47.0, starlette 1.3.1 (confirmed via pip show inside the image).
• route_spec.rb passes 6/6 via the real Docker + rspec harness (uvicorn image).

Related

• Supersedes PR Pin fastapi_startkit benchmark to fastapi-startkit==0.46.0 #4 (which pinned ==0.46.0).
• README perf framing updated separately on the PR README: fastapi vs fastapi-startkit comparison #7 branch to drop the now-resolved version-mismatch disclosure and attribute the POST cost to FastAPI 0.139's _IncludedRouter.