ci: fix live server-capture gate (runner lost NO_PROXY) + print server.log on failure

[maocheng23]

Motivation

The live server-capture gate in PR Test fails on every PR since 09-30, for example #934 (twice) and this PR's first run:

RuntimeError: sglang server did not become healthy (exit code -9); tail of .../server.log:
...
[...] Uvicorn running on http://127.0.0.1:30989
[...] Initialization failed. warmup error: ...
AssertionError: res=<Response [502]>, res.text=''
[...] kill_process_tree called: parent_pid=...

Root cause: the self-hosted runner still exports HTTP_PROXY/HTTPS_PROXY=http://172.17.0.1:7890 into the job container, but NO_PROXY/no_proxy are no longer exported. They were present in the last green run (09-30, NO_PROXY=localhost,127.0.0.1,::1,...) and are absent in every run since. sglang's startup warmup POSTs to 127.0.0.1, so the request goes to the proxy, which returns 502, and the server kills itself.

Changes

• ci: set NO_PROXY=localhost,127.0.0.1,::1 at the unit-test job level, so loopback traffic bypasses the proxy regardless of the runner's environment.
• test(capture-gate): when the server doesn't become healthy, include its exit code and the last 80 lines of server.log in the error. That's how this cause was found; before, the job log only gave a path inside the discarded container.

Runner admins may also want to restore NO_PROXY in the runner's own environment, since other loopback traffic in the job is affected too.

🤖 Generated with Claude Code