Sync latest GeneMamba docs and next-token training updates

Update Phase 3 to next-token train-from-scratch with checkpoint auto-resume

Use next-token training and checkpoint resume in train-from-scratch

Update continue pretraining docs and tokenizer details

Sync README model card from project root

Add GitHub dataset source link in README

Add model card YAML metadata

Upload GeneMamba model

Flatten examples: move downstream files to examples root

Unify repo: default 24l-512d at root, add size variants via subfolder