[5/9] additional_model_arguments

[garrett361]

Adds --additional_model_arguments for passing additional args to models (e.g. for z-loss or NoPE)

	PR	Title
1	#15	padding-free
2	#16	clean_checkpoints_at_end
3	#17	final_lr_ratio
4	#18	add_seed_and_date_to_run_name
5	>19	additional_model_arguments
6	#20	sync_each_batch=True grad acc
7	#21	no grad acc averaging for sum losses
8	#22	extra reporting
9	#23	local_main_process_first when building dataset

[garrett361]

@fabianlim we have this as a additional_model_arguments: Optional[list[str]] which is then turned into a dict via regex. I think we can just do additional_model_arguments: Optional[dict]?

An then like --additional_model_arguments '{"key": value}'? Not sure.

[garrett361]

also fixed an import that needs fixing before the tests can run