Conformance: Support Pod Port in Data Parallelism Tests

[danehans]

#1769 adds the GatewayFollowingEPPRoutingWithDataParallelism test which only verifies pod-level confinement (i.e., requests never leak to pods that EPP didn’t return). It does not assert which port (i.e. rank) within a selected pod handled the request. Currently, the echo server used for tests does not echo back the port used for the request. When this echo server has been updated or a different echo server is used that echos the request port, update the GatewayFollowingEPPRoutingWithDataParallelism to assert the request port so DP can be properly tested.

[danehans]

@robscott thoughts on adding this functionality to the echo server?

/cc @shmuelk @nirrozenbaum @zetxqx

[danehans]

llm-d/llm-d-inference-sim#232 added the pod port header and already supports headers for pod name/ns so it could be used as a replacement for the echo server.

[shmuelk]

Actually I am in favor of eliminating the echo server and using the llm-d-inference-sim instead