feat: update llama.cpp to 0c1e57098

[abetlen]

Update vendored llama.cpp from fb34fc262 to 0c1e57098.

• Bind the extended batch API, llama_process, and llama_get_causal_attn.
• Sync the MTMD post-decode callback with the new embedding-batch structure and update the unreleased changelog entry.
• Verified a local CPU build, all 26 Python tests, Ruff, extended-batch logits equivalence, and a real Qwen3.5 image encode/decode callback.