Replace model.safetensors with the canonical fine-tuned weights

The previous model.safetensors (sha 437b7ecef...) was structurally a Base
model copy: of 148 transformer-body tensors, 147 were bytewise-identical to
UCL-CSSB/PlasmidGPT, and only the embedding/lm_head pair differed. Anyone
loading this repo via AutoModelForCausalLM.from_pretrained got back
essentially Base, masking the real fine-tune.

The canonical weights were sitting under the non-standard filename
PlasmidGPT Finetuned Model.safetensors (sha 857d91938...). This commit
promotes those to the standard model.safetensors filename and removes the
non-standard duplicate.

End state: one model.safetensors in the repo, sha 857d91938..., containing
the real fine-tuned weights (148 of 148 layers differ from Base, mean
per-parameter diff 0.0037).

PlasmidGPT Finetuned Model.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:857d919382efb3e7564a9af08c7c35dd926b768d7b2b0a799ad587b589fe4111
-size 489030728

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:437b7ecef6996e7c8cc523b59c2aaa492f344e17320551f2b797369f50b881f7
+oid sha256:857d919382efb3e7564a9af08c7c35dd926b768d7b2b0a799ad587b589fe4111
 size 489030728