fix: normalize explicit TSQA choice labels

Track remaining evaluation bundles

Align caption and TSQA evaluation pipelines

ChatTS OOS runner: drop English datasets (cap_en, qa_en), keep zh caption + QA; remove redundant run_chatts_qazh.py

GPT inference: align TS serialization to text-LM peers + streaming/resumable

README: document the caption GLM-extraction step (metric key + statistical-property explanation)

Caption-scoring pipeline fixes + with-definitions GLM extractor

Add repository README

Add slimmed 4level_grpo_min2 + joint_grpo_min2 checkpoints; peer models -> placeholder

TS-Align benchmark reproduction bundles + canonical eval data + dataset sources