feat: add SimpleStorage payload transfer abstraction

[Ginray]

Summary

• add the narrow PayloadTransfer boundary for SimpleStorage large payloads
• keep ZMQ as the default payload path and control plane
• provide an optional UCX Tagged Host-memory implementation for RoCE RDMA
• preserve Mooncake, YuanRong, RayStore, and existing backend ownership
• use node-local GID/netdev and UCX capability discovery without machine-specific NIC names
• add the developer guide for official UCX installation, TQ build/use, and RDMA lane verification
• mark old PR-branch experiments as historical and link the current official-release result

Why

SimpleStorage needs a small data-plane boundary for large encoded payloads without changing its
ZMQ control protocol or forcing other storage backends onto the same transport. UCX is opt-in and
does not replace the storage backend or control plane.

User impact

The default remains:

payload_transfer: zmq

Users can select ucx when TQ is built with UCX support and each participating Ray node has a
matching RoCE-v2 path. The new developer guide explains the official UCX install, native build,
runtime configuration, and how to verify rc_verbs data versus TCP auxiliary traffic.

Validation

• ruff check . passed
• Python syntax checks passed
• git diff --check passed
• official, unmodified UCX v1.22.0 was used on A2-26/A2-27
• SimpleStorage cross-node UCX PUT/GET/CLEAR passed
• UCX logs showed tag(rc_verbs/hns_0:1) ka(tcp/enp189s0f0)
• manager throughput sample: 128 KiB 28.75/30.75, 1 MiB 56.03/73.07,
  16 MiB 96.78/97.02 MiB/s for PUT/GET

The current native close path can leave a server-side Connection reset by remote peer warning
after the client force-closes; the data and CLEAR checks still pass.

[Ginray]

Superseded by the consolidated implementation PR #6 from ucx-payload-o1.