fix distill

release small navit perf

address #352

fix positional embed for mean pool case and cleanup

address #351

VAAT should have two ears

register tokens for the AST in VAAT

improvise a variant of VAT with audio cortex before fully generalizin…

…g it

the paper claims finetuning everything was better, but just allow for…

… freezing the visual cortex, what PI proposes