[serve] Enable gRPC streaming tests under HAProxy

[eicherseiji]

Why

HAProxy proxies gRPC end to end over HTTP/2. The gRPC frontend bind and the backend server lines both carry proto h2, so HAProxy decodes and forwards HTTP/2 streams, including client-streaming and bidirectional streaming, without any extra handling. gRPC-over-HAProxy support landed in #63735, and the streaming cases were skipped at that time as a scope cut rather than because of a real limitation.

What changed

1. Remove the skipif(RAY_SERVE_ENABLE_HA_PROXY, reason="HAProxy does not support streaming.") markers from the 13 gRPC streaming tests in test_grpc.py.
2. Run the Streaming method assertion unconditionally in test_serving_grpc_requests (test_grpc.py) and test_serving_request_through_grpc_proxy (test_cli_4.py). Each previously gated that assertion behind an inline conditional (if not RAY_SERVE_ENABLE_HA_PROXY and if not RAY_SERVE_ENABLE_DIRECT_INGRESS). HAProxy forces RAY_SERVE_ENABLE_DIRECT_INGRESS on (constants.py), so both conditions were false under HAProxy and the assertion never ran there. Also drops the now-unused RAY_SERVE_ENABLE_HA_PROXY import in test_grpc.py.

The serve: HAProxy tests CI step runs the full serve suite under RAY_SERVE_ENABLE_HA_PROXY=1, so these all execute there now instead of being skipped.

Testing

Ran under the bundled HAProxy 2.8.25 with RAY_SERVE_ENABLE_HA_PROXY=1 RAY_SERVE_ENABLE_DIRECT_INGRESS=0:

python -m pytest python/ray/serve/tests/test_grpc.py -k streaming
# 20 passed, 23 deselected

python -m pytest \
  python/ray/serve/tests/test_grpc.py::test_serving_grpc_requests \
  python/ray/serve/tests/test_cli_4.py::test_serving_request_through_grpc_proxy
# 2 passed

Streaming coverage now exercised under HAProxy: client and bidirectional streaming, the grpc-context variants, timeout, cancellation, backpressure, mid-stream client errors, and the server-streaming Streaming method in the gRPC and CLI end-to-end tests.

Not a duplicate

No open PR or issue targets gRPC streaming under HAProxy (checked gh pr list and gh issue list on ray-project/ray).

[gemini-code-assist]

Code Review

This pull request removes the RAY_SERVE_ENABLE_HA_PROXY constant and the associated conditional skips for gRPC streaming tests in test_grpc.py and test_cli_4.py, allowing these tests to run unconditionally. There are no review comments to evaluate, so I have no additional feedback to provide.