perf: disable llama perf counters in hot path

crates/inference-core/src/engine.rs

@@ -232,7 +232,7 @@ impl InferenceEngine {
             LLAMA_FLASH_ATTN_TYPE_DISABLED
         };
         cparams.offload_kqv = true;
-        cparams.no_perf = false;
+        cparams.no_perf = true;
 
         // 6. Init context
         let ctx_ptr = unsafe { (lib.init_from_model)(model_ptr, cparams) };

crates/inference-core/src/session.rs

@@ -112,7 +112,8 @@ impl GenerationSession {
         debug!(n_tokens = n_prompt_tokens, "Prompt tokenised");
 
         // ── 2. Build sampler chain ─────────────────────────────────────────
-        let sc_params = unsafe { (lib.sampler_chain_default_params)() };
+        let mut sc_params = unsafe { (lib.sampler_chain_default_params)() };
+        sc_params.no_perf = true;
         let sampler_chain = unsafe { (lib.sampler_chain_init)(sc_params) };
         if sampler_chain.is_null() {
             bail!("sampler_chain_init returned NULL");