feat(coordinator) Add DP-rank pinning parity to the coordinator's kv-nixl connector

[roytman]

What type of PR is this?
/kind feature

What this PR does / why we need it:
The sidecar's nixlv2 KV connector pins a disaggregated request's prefill and decode legs to the same vLLM data-parallel rank via the x-data-parallel-rank header, guarding against SO_REUSEPORT-sharing DP backends routing the two legs to different ranks. The coordinator's equivalent connector, kv-nixl, had no DP awareness at all.

This adds the same behavior to the coordinator:

New pipeline.dp_size config option (default 1, disabled), merged into step params the same way kv_connector/ec_connector already are.
kv-nixl hashes the request ID into a deterministic rank and sets it as the x-data-parallel-rank header on both the prefill and decode requests, preferring a remote_dp_rank the prefill response echoes back when present. kv-sglang and kv-shared-storage are unaffected — their sidecar equivalents never reference DP rank either.
No kv_transfer_params body field changes; this is header-only, matching the sidecar's plain (non-MoRI-IO) NIXLv2 behavior.

Which issue(s) this PR fixes:

Fixes #2478

Release note (write NONE if no user-facing change):

Add `pipeline.dp_size` coordinator config option for the `kv-nixl` connector: pins a request's prefill and decode calls to the same vLLM data-parallel rank via the `x-data-parallel-rank` header, matching the llm-d-router sidecar's `nixlv2` connector behavior.