The nullifier algorithm was wrongly assuming that it would received an
ordered sequence of transactions. When combined with the cancelation
search only looking forward, this could cause the algorithm to not only
not throw the `MoreThanOneMatchError`, but also attempt to rewrite a
previously matches transaction.
The fix is applied both in ordering every sequence of transactions, but
also to confirm that a transactions had not previously been found. This
should now be impossible given the correct order, but its there to
prevent future misshaps.
For the specific broken sequence, look at test
TestTransform.test_nullifier_inplace_unordered.
The FK are correctly initialized when the rules are created as part of
the base category/tag.
Also removes the name from the CategorySchedule, same logic applies.
The `CategorySelector` was possibly added to be incremented with other
attributes. However, since none other that the selector enum is used at
the moment, it is only adding unnecessary cluter.
The category selector value is moved to the parent
`TransactionCategory`.
Due to the use of the dataclasses mixin on the SQLAlchemy types, a
back_populates creates a RecursiveError when comparing two types. This
occurs because the dataclass will overwrite the __eq__ operator, and it
doesn't know when to stop comparing relationships.
Removing the dataclasses isn't the best approach, since then __init__,
__eq__ and __repr__ methods would have to be added to all types. Thus
the solution was to remove the relationship on the child (on a
one-to-one relationship) from the __eq__ operation, with the use of the
compare parameter.
Took the opportunity to define more logical __init__ methods on the
`Rule` and child classes.
Also revised the parameter options on some DB types.
