Match Qwen reasoning history across effort changes

docs/issue117-reasoning-history/reference.log

@@ -0,0 +1,5 @@
+[transformers] You are using a model of type `qwen4_exp` to instantiate a model of type ``. This may be expected if you are loading a checkpoint that shares a subset of the architecture (e.g., loading a `sam2_video` checkpoint into `Sam2Model`), but is otherwise not supported and can yield errors. Please verify that the checkpoint is compatible with the model you are instantiating.
+[tokenizer] restored the pre-tokenizer regex from tokenizer.json: the loaded tokenizer split combining marks off their letters (Devanagari, Thai, vowelled Arabic tokenized 25-90% longer than the model was trained on)
+[mtplx] chat turn segmentation {"enabled": true, "reason": "structure holds and all 12 probe renders give the single-call ids when cut", "tokenizer": "TokenizerWrapper"}
+{"cases": 304, "direct_example_tokens": 59, "legacy_direct_history_tokens": 40}
+{"mode_switches": [{"from_effort": "Low", "direct_context_tokens": 62, "exact": true}, {"from_effort": "Medium", "direct_context_tokens": 62, "exact": true}, {"from_effort": "XHigh", "direct_context_tokens": 62, "exact": true}]}