* Remove the no-op rmtree guard around the GGUF save
patch_unsloth_gguf_save saves shutil.rmtree and restores it, but never
replaces it, so the context manager does nothing. Its comment claims it
prevents deletion of the directory save_pretrained just created.
It reads as a copy of the patch_unsloth_save sibling above it, minus the one
line that does the work. Completing it is not the right fix though: the GGUF
call forces push_to_hub=False and the merge cleanup that would remove the save
directory is gated on push_to_hub=True, so nothing on this path calls rmtree.
That was verified on a real LoRA-backed q8_0 export with every rmtree call
logged, in the discussion on #7149.
Drop the dead context manager rather than leave code that looks like a guard
and is not. Behaviour is unchanged; the following step comments are renumbered
to stay contiguous.
* Note why no rmtree guard is needed at the GGUF call site
---------
Co-authored-by: danielhanchen <michaelhan2050@gmail.com>
274f5ff569