A missing pair of message delimiters made every DeepSeek V4 coding-agent turn re-prefill 12,651 tokens.

llama.cpp PR #29008, merged September 17, adds <｜User｜> and <｜Assistant｜> as message delimiters to the DeepSeek V3.2 parser, which an earlier PR had skipped when adding them everywhere else. Because V3.2/V4/V4.1 use SWA layers and the KV cache can't roll back to a position without a checkpoint, llama-server only placed checkpoints at end - 4 and end - 4 - n_ubatch, so any divergence before the tail re-prefilled from token 0. On DeepSeek-V4.1-Flash on one RTX A6000 with a 13,167-token agent prompt, a new user question went from 510 prefilled tokens in 13.8s to 16 tokens in 0.8s; editing the second of three user turns went from a full re-prefill to 79 tokens in 5.0s. Two missing strings, a 17x latency difference.