Randomize candidate dataset order

Fix: No pairs are created for mixed queries

fix: mixed templates are not generated because of stale parquet files

chore: clean up the file references, remove unused scripts for gemma finetuning

Add a couple of more sql templates for natural earth & chained queries

Add finetune instructions for gemma

ENH: Add dataset creation on scale using modal clusters

FEAT: Add to create SLM training data