Merciful validation

[ivanov]

This is related to #235 but provides for an alternative path. There are instances when users would want to fixup the notebook, so we should provide a way for them to do so.

Right now, if you have a validation error for non-unique cells, you will get a validation error informing you that a cell had a duplicate cell_id, and also the notebook that you passed in will be modified with only the first instance of a non-unique cell. I agree with the approach @Carreau outlines in #235, but want it to be possible to opt-in to fixing the validation errors that are fixable (such as replacing all non-unique cell ids with unique ones). I think it would makes sense to do this via a new repair kwarg that would default to False, and keep with either current behavior, or non-modifying behavior as per #235, but when repair=True we would fix up the errors and do a best effort of massaging the notebook into one that is compliant with a given nbformat version.

[Carreau]

Sorry for the late reply, the problem with the kwarg, and the approach of fixing in validate is the following.

notebook = get_notebook()
notebook_backup = deepcopy(notebook)
do_stuff(notebook) # like for example compute signature.
validate(notebook)
assert notebook_backup == notebook

This is an assumption many codebase do, that validate is not mutating notebook, and that validate()/do_stuff() can be reordered. And no keyword are going to fix that.