fix: remove sharded safetensors — breaks Ollama GGUF detection

Ollama's repo type detection sees sharded safetensors (model-*-of-*.safetensors)
and classifies the repo as transformers, ignoring the GGUF files. Single-file
safetensors repos (like lemer) work fine — it's the shard index that triggers it.

BF16 safetensors for vLLM/transformers users remain available via the MLX
sibling repos or will be restored on a dedicated branch when Google ships
final Gemma 4 weights with MTP.

Co-Authored-By: Virgil <virgil@lethean.io>

model-00001-of-00002.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:1127684971bbca40465435a5cad69d67ad603bf5e61c6dfd5561fae4a3bcfdb3
-size 49907246508

model-00002-of-00002.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:aab47033e1e8a492ef8e581efae1cf36478d0433567e7729b3c1728bc8970db7
-size 1704763408