docs: drop thinking note per review

Author: kashif

docs/speculative.md

@@ -33,9 +33,7 @@ llama-server -m Qwen3-4B.gguf -md Qwen3-4B-DFlash.gguf \
     --spec-type draft-dflash --spec-draft-n-max 15 -fa on --jinja
 ```
 
-DFlash expects chat-formatted input (`--jinja`); setting `LLAMA_SPEC_NO_THINK=1` to disable
-"thinking" substantially raises the acceptance rate. `--spec-draft-n-max` is clamped to the draft
-model's trained block size.
+`--spec-draft-n-max` is clamped to the draft model's trained block size.
 
 See: