Reuse CP shared KV remaps across layer materialization

CP shared KV materialization repeatedly rebuilt the same logical-page slot remaps and page inverse metadata for each layer. Cache the token and paged remap metadata on the forward batch so MLA KV, index K/scale, and prefetch paths can reuse the layer-independent mapping while still materializing layer-specific data through the existing tai/torch runtime paths.

Constraint: Only mapping metadata is batch-scoped; dense KV/index contents remain layer-specific and are not reused.
Rejected: Cache fully materialized dense KV/index buffers | would add large per-layer memory residency and invalidation complexity.
Confidence: medium
Scope-risk: moderate
Directive: Do not assume this removes materialize or CP all-reduce cost; profile tai fallback logs and Nsight kernels before attributing E2E gains or losses.
Tested: git diff --check
Tested: remote g0034 container PYTHONPATH=python python3 -m pytest test/registered/unit/mem_cache/test_cp_shared_kv_runtime.py -q (52 passed, 5 warnings)
Not-tested: Full GLM-5 disaggregated E2E performance run

docs/advanced_features/nsa_prefill_cp_phase8_mla_prefix_prefetch_plan.md

@@ -18,6 +18,14 @@ Layer L attention 返回时不强制等待 prefetch
 Layer L+1 consume prefetched buffer 时 wait event，再补齐 suffix/current pages
 ```
 
+后续在 HiCache 分支补充了 batch-scoped slot remap 复用：`metadata.real_page_table`
+和 index page table 在同一个 forward 内跨 layer 不变，因此只在第一次 materialize /
+prefetch 时构造 `slot_logical_pages`、`page_inverse`、paged `dense_pages`，后续 layer
+复用该计划；每层仍重新 materialize 本层 KV/index 数据并执行 CP all-reduce。这个优化
+只复用“逻辑页到 dense slot 的映射计划”，不复用 dense KV 内容，避免跨层 KV 数据串用。
+如果同一个 `ForwardBatch` 已经带有 remap cache 但无法复用（key mismatch 或 cache
+状态不完整），运行时会打印 rate-limited warning；第一次冷启动构造 cache 不打日志。
+
 保留的环境变量：
 
 ```text