ENH: Support masked ExtensionTypes when using `engine="numba"`

[mroeschke]

Currently any engine="numba" operation is only compatible with numpy dtypes as numba does not natively understand pandas ExtensionDtypes. To achieve this, pandas could

1. Always convert to the equivalent numpy dtype, but operations with NA would likely need more logic.
2. Register the types with numba? https://numba.pydata.org/numba-doc/latest/extending/low-level.html#typing