Ensure torch 2.6+ compatibility, fix chain import and removal order in prune_program

[Meganton]

PyTorch 2.6+ changed the default behaviour of torch.load() to weights_only=True. This blocks loading full model objects by default for security reasons.
Setting weights_only to False will load the entire model, bypassing the new restriction.
Source: https://medium.com/@roscoe.kerby/fixing-the-weights-only-load-failed-error-in-pytorch-9098d0a44a9a

Also adding from itertools import chain, as it is needed to import and run prune_program.

Also when removing redundant lines in prune_program, converting the set to a list did not sort it, meaning it could happen that earlier lines were removed first and then later lines became mis-indexed, leading to wrongly deleted lines, or out of range-errors.