fix(tests): de-flake serve tests by isolating ports and failing fast

[bhimrazy]

What does this PR do?

Fixes the recurring flake in tests/test_serve.py, most often seen on test_serve_with_generate_strategy[tensor_parallel]:

AssertionError: Server did not respond as expected. 
Error: HTTPConnectionPool(host='127.0.0.1', port=8000):
Max retries exceeded ... Connection refused

Each test now serves on its own OS-assigned free port, so servers can no longer collide:

• Waiting stops as soon as the server process exits, and reports its exit code plus a tail of the server log instead of a bare connection error — e.g. Server exited with code 2 before it was ready. right away, rather than a connection refused after 30s.
• Teardown always runs through a context manager and waits for the whole process tree to be gone before the next test starts.
• Server output goes to a file rather than an unread pipe, which can fill up and block a chatty server (NCCL_DEBUG=INFO on CI) before it binds.
• The boilerplate duplicated across all seven tests collapses into two helpers.

Root cause

All seven tests bound the hardcoded port 8000, and teardown never waited for the server to actually exit — kill_process_tree only sends the signals, and the thread wrapping the non-blocking Popen was joined instead of the process. Three compounding problems:

• Port still held. A server that has not finished shutting down makes the next one exit at bind_socket() via uvicorn's silent sys.exit(1), which the test can only observe as a connection refused.
• Leaked servers. When an assertion fails, the kill_process_tree call below it is never reached, so the server survives for the rest of the session and keeps port 8000. tensor_parallel runs last, so it inherits the debris.
• No diagnostics. Nothing checked whether the server was still alive, so a crash at second one looked exactly like a slow start.