fix: benchmarks measured 307 redirects instead of the endpoints

[Vladgasik]

Situation

The repository publishes throughput comparisons between FastAPI configurations,
measured with ab against containers in docker-compose.yml. The endpoints
under test are declared with a trailing slash (/sync/big_json_response/,
/async/items/)

Complication

33 of those measurement configurations request the endpoints without the
trailing slash. FastAPI's default redirect_slashes=True answers 307 with
empty body and ab doesn't follow redirects. So those runs timed the emission
of a redirect, not the endpoint.

Resolution

Merge this branch (no new dependencies)

---

Evidence

The mismatch

@app.post("/sync/big_json_response/")          # app_files/app.py:63
container["uri"] = "/sync/big_json_response"   # test_files/test_combo_best_worst.py:20

Against the unmodified combo containers:

$ curl -s -o /dev/null -w '%{http_code} %{size_download}\n' -X POST \
    -H 'Content-Type: application/json' -d @test_files/requestbody \
    http://127.0.0.1:8131/sync/big_json_response
307 0

$ curl -s -o /dev/null -w '%{http_code} %{size_download}\n' -X POST \
    -H 'Content-Type: application/json' -d @test_files/requestbody \
    http://127.0.0.1:8131/sync/big_json_response/
200 916852

What each URL measures

Harness parameters (ab -c 100 -n 500), driven from a container inside the
compose network, warm-up plus 3 runs averaged:

Container	URL	RPS	non-2xx
combo_worst_sync_json	/async/big_json_response	4519.4	1500 / 1500
combo_worst_sync_json	/async/big_json_response/	21.6	0
combo_best_async_orjson	/async/big_json_response	17182.7	1500 / 1500
combo_best_async_orjson	/async/big_json_response/	46.1	0

The published 3433 RPS on a 1 MB payload would require ~3.4 GB/s of serialisation
and transfer inside a 2-CPU container.

The guard

AssertionError: 200 non-2xx responses from /sync/big_json_response on port 8131.
The measurement is not hitting the endpoint: FastAPI answers 307 when the
trailing slash of the route is missing from the URL.

test_pool_exhaustion, test_robustness and test_concurrency drive the app
through httpx rather than TestContainer, so the assertion does not affect them.
The middlewares, json_classes, sync_async and server_runners tests already
used correct URLs and are unchanged by this PR.

Deliberately out of scope

The measurement tables in index.md, nginx_port_socket.md,
workers_and_threads.md and keepalive.md came from the broken configurations
and need a fresh CI run on your hardware; numbers from a different machine would
not be comparable, so they are left untouched.

Expect a longer suite: the big-JSON tests previously ran at redirect speed
(~4500 RPS) and now run at endpoint speed (tens of RPS).

[KissPeter]

Thanks @Vladgasik, let me check it