Mihaiii/Pallas-0.5-frankenmerge

Same weights, repacked vs Mihaiii/Pallas-0.5. Same weights in different files (re-saved, re-sharded or new metadata), plus 27 new tensors.

Tensors changed: Embeddings 0/1, Norms 0/121, MLP 0/180, Attention 0/240, lm_head 0/1

2 derivatives checked