test(config): assert aiohttp User-Agent test actually verifies the request

[anxkhn]

test_agent_is_used_aiohttp in tests/test_config.py exercised the aiohttp
User-Agent override (_config.set_http_lib_agents({"aiohttp": ...})) but threw
the response away (html = await response.text()) and had no assertion, so it
verified nothing. A regression in the aiohttp branch of set_http_lib_agents
would still pass this test, i.e. it is a false-coverage / mutation-survivor test.

Its two siblings both assert the request was actually served with the expected
header:

• test_agent_is_used_requests: assert requests.get(...).status_code == 200
• test_agent_is_used_httpx: assert httpx.get(...).status_code == 200

This makes the aiohttp coroutine return response.status and asserts it equals
200, mirroring the siblings. pytest_httpserver only serves the request when
the incoming User-Agent matches the expected header, so if the aiohttp
User-Agent override breaks, the request goes unserved (500) and the test now
fails instead of passing silently. A terse assert message documents the expected
behaviour, as the repo's contribution guide asks.

Test-only, one file, +4/-2. No production code changes.

Verification

• Run the tests and ensure they pass python -m pytest tests/
  • python -m pytest tests/test_config.py::test_agent_is_used_aiohttp -> 1 passed
  • python -m pytest tests/test_config.py -> 71 passed (no regressions)
• Verify the test now fails when the behaviour it guards is broken:
  temporarily disabling the aiohttp branch of set_http_lib_agents makes
  pytest_httpserver return 500 for the unexpected User-Agent, and the new
  assertion fails (assert 500 == 200) with the descriptive message. It passes
  again once the override is restored. This confirms the test is now a real
  red/green regression guard rather than a vacuous pass.
• Formatting unchanged: black --check tests/test_config.py.

Not a duplicate

No open PR touches test_agent_is_used_aiohttp or set_http_lib_agents in a
related way. The open PRs that edit tests/test_config.py do so at unrelated
functions: #1857 (deprecated -m/--model_* CLI-flag removal), #1816 (a
parallel-attempts test), #1790 (structured exit codes; its set_http_lib_agents
calls save/restore prior agents inside the worker-count test, not this one), and
#1651 (exit-code tests). A keyword scan of open PRs for aiohttp / user-agent
/ set_http_lib_agents / agent is used returns nothing relevant, and no open
issue tracks this test. (The related leakreplay test-coverage fix is a separate
PR, #1914, which touches only tests/probes/test_probes_leakreplay.py.)

AI assistance

AI assistance was used to prepare this change. I have reviewed every changed
line, reproduced the missing-coverage behaviour and the fix locally (confirming
red -> green), and ran the test suite for the affected file.

---

Notes for maintainers scope

• Follows the test(config): conventional-commit style already used in the repo
  (e.g. test(config): regression guard for config_files dedup #1791).
• DCO sign-off and a Co-authored-by: Claude trailer are in the commit.
• No new dependencies, no base-class edits, no SPDX header (nothing imported
  under another license).