llama.cpp slot restore still reprocesses prompts

This looks less like broken persistence and more like two different mechanisms getting conflated: disk-backed slot state is being saved, but runtime prefix/KV reuse still has its own eligibility checks.

• –`--slot-save-path` enables `/slots/{id}?action=save|restore` and stores the slot's prompt cache to disk, not a generic “resume everything” checkpoint.
• –`n_written` is the serialized slot payload size, so a large file is expected for long contexts; it does not automatically mean the model will skip re-prefill on restore.
• –The `cache reuse is not supported` and `forcing full prompt re-processing due to lack of cache data` logs point to a separate reuse path being rejected at runtime.
• –Similar upstream Qwen3.5 reports show the same behavior, which makes this look like a model/runtime limitation or bug rather than a missing command-line flag.
• –`--swa-full` alone is not enough to guarantee resumable KV reuse for hybrid/SWA-style architectures.