expose dataloader configuration in all configs

[felipemello1]

Currently, we dont allow users to decide on their own the number of works, pin_to_memory, etc. We should expose it in the configs.

Recipe example: https://github.com/pytorch/torchtune/blob/7bfb3336446f0d874ab5d4595249839b735b7076/recipes/full_finetune_single_device.py#L578

then in the config we have something like:

dataloader:
	batch_size:  ...
	num_workers: 0 # set to -1 to use all. This may not work in some setups.
	pin_memory: False # Set to true to xyz
	#TODO: check if it makes sense to expose other arguments

The script in this PR should help to bulk update the configs: #1954

[RdoubleA]

cc @andrewkho, we'll likely have a dataloader builder for the torchdata components, and will have to expose this in the config

[andrewkho]

makes sense, definitely in the plan