Stop calling astype('Float64') after `replace` call once Pandas bug is fixed

Had to add a call to convert the dtype back to `Float64` after a `DataFrame.replace` call because of a Pandas bug.

https://github.com/alteryx/woodwork/pull/755/files#diff-75cb54847db4ed09b32148b93f1f543df16a421473154cb17085ff4932277ba8R1140

It should be removed once the bug is fixed and the `replace` call doesn't change the dtype.
- https://github.com/pandas-dev/pandas/issues/40732