docs: drop underpowered n=20 in-distribution table, rely on Spider for SQL capability

docs: fix in-distribution results table, correct comparison claim, add training details

Update quick-start snippet to reference renamed repo

Fix license metadata (DeepSeek Model License, not apache-2.0) and add dataset credits

Rewrite model card: correct repo references, mixed-training results, intended use & limitations

Replace with SQL+Magicoder mixed-training adapter (better SQL generalization + retained code capability, see experiment_log.md)

Update adapter: mix Magicoder code data (30% SQL / 70% code) to mitigate code-capability degradation

Add LoRA adapter

initial commit