GeezerVison

Fixes #504

Problem

The fastapi benchmark (bm_fastapi) fails when the environment has an HTTP
proxy configured — e.g. HTTP_PROXY/http_proxy set, which is common in CI or
corporate environments — and 127.0.0.1 is not listed in NO_PROXY.

The benchmark starts a uvicorn server on 127.0.0.1 and then talks to it with
httpx.AsyncClient(). Because httpx reads proxy configuration from the
environment by default, the request to http://127.0.0.1:<port>/items/0 is
routed through the proxy. The proxy cannot reach the benchmark machine's
loopback address, so it returns 503 Service Unavailable and the benchmark
dies:

httpx.HTTPStatusError: Server error '503 Service Unavailable' for url 'http://127.0.0.1:40027/items/0'
...
RuntimeError: Benchmark died
ERROR: Benchmark fastapi failed: Benchmark died

Fix

Create the client with trust_env=False so it never reads proxy (or .netrc)
settings from the environment:

async with httpx.AsyncClient(trust_env=False) as client:

This is correct for this benchmark because it only ever connects to its own
loopback server, so it must never be routed through a proxy. In environments
with no proxy configured, trust_env=False behaves identically to the default,
so benchmark results are unaffected for the common case.

Only bm_fastapi is affected: the other networking benchmarks
(bm_tornado_http, bm_asyncio_tcp, bm_asyncio_websockets, bm_async_tree)
use clients that do not read proxy environment variables.

Testing

Reproduced by running the benchmark with proxy environment variables inherited
and 127.0.0.1 absent from NO_PROXY:

$ python -m pyperformance run bm_fastapi --inherit-environ http_proxy,https_proxy,no_proxy

• Before the change: Benchmark fastapi failed: Benchmark died with the 503 above.
• After the change: the benchmark completes and reports a result.