Changes Partition to optionally take an input dtype

[petim0]

I changed partition to allow to optionally pass a dtype for the input register.

This change comes from a bug I encountered when partitioning a QInt and running assert_consistent_classical_action:
The partition assumes that your input is a QAny() and QAny() is assumed to have a _UInt encoding, therefore when the _classical_partition function of _PartitionBase uses self.lumped_dtype.to_bits(x) and you have a negative number as a input then you get an error you should not have, as the _UInt encoding doesn't allow for negative numbers.

Mypy was really not happy with the match statement and so I had do do a lot of useless asserts. Maybe a better solution is I to just ignore the errors instead.

I also added a LegacyPartitionWarning that I then put in the filterwarnings of pytest as to not have it everywhere in the code when testing.

I also choose to have the base dtype of the partition register to be a QUInt as to more reflect the fact that we assume a _Uint encoding.

[petim0]

This technically solves #1799, but I haven't done any tests so I am not sure. Also maybe still doing a classical partition is not a bad idea.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[mpharrigan]

main philosophical discussion on #1813

Yes, I think we should let you partition any data type so you can write safe, generic quantum programs.

The constructor for Partition has to be backwards compatible and I don't think we should deprecate partitioning a QAny. You can monkey with the order and defaults of the constructor to make a backwards-compatible way of accepting a dtype argument or, my preference:

I think this would work better as a new PartitionT class that leaves legacy Partition alone and uses the fact that the logic is now factored out into a base class to ones advantage. You can add a "see also / consider instead" to the legacy partition docstring