fix(rdma): align server/client depth, max_msg and MAX_BLOCK_BYTES defaults

[ketor]

Problem

On GLM-5.2-NVFP4 (DSA multi-pool model) with dfkv L3, the hot round throughput is -29.8% worse than cold when server/client RDMA defaults are used without explicit configuration:

Metric	Cold	Hot	Change
Total throughput	37,306 tok/s	26,191 tok/s	-29.8%
Median TTFT	15.1s	32.4s	+114%
Prefetch failures	0	532	---
Depth clamp warnings	>0	---	---

Root Cause

Server and client RDMA pipeline depth default=1 while production uses depth=4. The mismatch causes batching window clamp, PUT batch failures during burst writes (128 pages per batch), and subsequent hot-round prefetch failures.

Additionally, client DFKV_RDMA_MAX_BLOCK_BYTES defaults to 64 MiB (inherited from max_payload), allocating ~256 MiB per connection and capping receive segment capacity at ~256 connections. GLM-5.2-NVFP4 per-SG-chunk payload is ~1 MiB, so 64 MiB is 64x oversized.

Fix

Four default value changes:

1. Server RDMA depth 1 to 4 (rdma_server.cc)
2. Client RDMA depth 1 to 4 (rdma_transport.cc)
3. Client DFKV_RDMA_MAX_BLOCK_BYTES 64 MiB to 4 MiB (rdma_transport.cc)
4. Server --max-msg 64 MiB to 32 MiB (dfkv_server_main.cc)

Verification (xb01-0064, 8xB200, GLM-5.2-NVFP4, dfkv v2.6.3)

Before fix (server depth=1, client depth=4)

Metric	Cold	Hot	Hot vs Cold
Total throughput	37,306 tok/s	26,191 tok/s	-29.8%
Median TTFT	15,139 ms	32,448 ms	+114%
Prefetch failures	0	532	---

After fix (server depth=4, client depth=4)

Metric	Cold	Hot	Hot vs Cold
Total throughput	38,549 tok/s	37,409 tok/s	-2.9%
Median TTFT	13,940 ms	14,990 ms	+7.2%
Prefetch failures	0	0	---
Depth clamp warnings	0	0	---

Improvement

• Hot vs cold degradation: -29.8% to -2.9% (90% reduction)
• Prefetch failures: 532 to 0
• Hot throughput: +42.9% (26,191 to 37,409 tok/s)

Connection memory impact

Config	Per-conn (depth 4)	16 GiB segment capacity
Old: client 64 MiB	256 MiB	~64
New: client 4 MiB	16 MiB	~1024

Files Changed

• src/cache/rdma_server.cc: ServerDepth() default 1 to 4
• src/cache/rdma_server.h: header comment
• src/cache/dfkv_server_main.cc: --max-msg default 64 to 32 MiB, --rdma-depth help, --max-msg help
• src/transport/rdma_transport.cc: depth_ default 1 to 4, DFKV_RDMA_MAX_BLOCK_BYTES default 64 to 4 MiB
• src/transport/rdma_transport.h: header comments

Scope

• Changes defaults only; explicit env/flag still overrides
• No protocol change, no wire format change
• Existing deployments with explicit DFKV_RDMA_DEPTH=4 are unaffected