Fix export of all quantized transformer models

[eldarkurtic]

Our current transformers export-to-onnx pipeline doesn't work for quantized models. To reproduce: try to export any quantized model with the latest versions of sparseml and neuralmagic/transformers.

There are two reasons:

1. we are passing in resolved_archive_file=[] at

   sparseml/src/sparseml/transformers/sparsification/trainer.py

   Line 690 in b73a173

   resolved_archive_file=[],

   which causes error in HF library here: https://github.com/neuralmagic/transformers/blob/0798c9e3b743a7e5c552f943a1a7d52ff63bbffb/src/transformers/modeling_utils.py#L3150 . We can fix this by passing None instead of [].
2. the function self.model._load_pretrained_model from transformers returns 6 items (see https://github.com/neuralmagic/transformers/blob/0798c9e3b743a7e5c552f943a1a7d52ff63bbffb/src/transformers/modeling_utils.py#L3323), and in our interface we only accept 5

   sparseml/src/sparseml/transformers/sparsification/trainer.py

   Line 686 in b73a173

   _, missing, unexpected, _, _ = self.model._load_pretrained_model(

   which also raises an issue while unpacking the returned tuple from their function.

[bfineran]

great catch @eldarkurtic this should also fix our broken tests on main. Thank you!

[eldarkurtic]

Accidentally pushed changes from this PR (#1630) as well to enable --trust_remote_code, so we can close that PR by merging this one into main.

[bfineran]

failing tests are unrelated - investigating separately but unable to reproduce outside of GHA. Merging this PR to unblock functionality