nx_nofloat_serve_gate.nx

nx_nofloat_serve_gate.nx -- gates the no-float LLM serve CORE in-process (pure-core+gate+daemon idiom: no sockets here; synthetic HTTP requests straight into nsv_handle). Teeth: T1 GET /health -> 200 + ok:1 (init fail-fast worked; model + both caches live) T2 POST /gen "The capital of France is" (i32) -> 200 + READABLE " Paris. It is the" (real spaces -- the byte-level-BPE decode; raw pieces would read "GParis." with G-dot 0xC4 0xA0) T3 arbitrary second prompt -> 200 + gen_tokens>0 + NO 0xC4 0xA0 byte pair anywhere (decode total) T4 mode i8 on the France prompt -> 200 + contains "Paris" (fast mode serves) T5 GET /nope -> 404 (router honest) T8-T10 (2026-09-17, the prompt-length class, four pillars): nsv_prompt_admit refuses by BYTES, tk_bpe_encode_b refuses by capacity, and a prompt past the byte cap (T10a) or the token cap (T10b, more ids than NSV_MAXT held by the byte-sized ids buffer) answers 400 err 3 with the process alive; the pre-fix engine died in tk_rank_fast on both Requires /home/elderwesto/nx_stage/nx_real_model.gguf. expect_exit: 0 license_tier: ORIGINAL