vLLM cuts DeepSeek-V4.1's KV record from 584 bytes per token to 528.

PR #56893, merged September 15, switches V4.1 off the V4 paged fp8_ds_mla record onto a V4.1-specific one that quantizes the RoPE dims too: 512 B of fp8 e4m3 plus 16 UE8M0 scales, one per 32 dims, against V4's 448 B fp8 NoPE plus 128 B bf16 RoPE and 7 scales (GitHub). Because block_size is always a multiple of 32, every V4.1 page is an exact multiple of the 512 B TMA stride, so V4's padding disappears on top of the 9.6% smaller record. The debugging story is the good part: the V4 indexer cache's 576 B alignment existed only to keep the summed block stride divisible, and leaving it while the main caches moved produced a stride of 212160 that tripped a FlashMLA assertion after a clean weight load, with every per-spec page size correct in isolation.