add option to exclude specific indices from the dataset

[FloWsnr]

Currently, we can restrict the number of samples/trajectories for each dataset using the restrict_num_samples or restrict_num_trajectories parameters.

If you agree, I'd like to enable an option to pass a list of indices to skip as a third option in the _build_restriction_set method.

That way, we can skip faulty or useless samples. On example are the steady-state solutions in the gray-scott dataset, which might distort training. Now we can just precompute/list the the indices to remove.