Add EXO_DISABLE_LOGPROBS=1 to skip per-token logprobs extraction

For profiling: extract_top_logprobs() does 11 .item() calls +
argpartition on 248K vocab per token. Testing if this is the
source of speculative overhead vs mlx_bench.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

src/exo/worker/engines/mlx/generator/batch_generate.py

@@ -470,7 +470,7 @@ class ExoBatchGenerator:
 
             logprob: float | None = None
             top_logprobs: list[TopLogprobItem] | None = None
-            if task_params.logprobs:
+            if task_params.logprobs and os.environ.get("EXO_DISABLE_LOGPROBS") != "1":
                 logprob, top_logprobs = extract_top_logprobs(
                     logprobs=response.logprobs,
                     tokenizer=self.tokenizer,