REF: use _unbox_scalar to de-duplicate TDA/DTA/PA comparison methods

[jbrockmendel]

DatetimeArray and TimedeltaArray comparison methods share a lot of code, might be possible to de-duplicate a lot of it by using the new unbox_scalar method:

From DTA:

        if isinstance(other, (datetime, np.datetime64, compat.string_types)):
            if isinstance(other, (datetime, np.datetime64)):
                # GH#18435 strings get a pass from tzawareness compat
                self._assert_tzawareness_compat(other)

            try:
                other = _to_m8(other, tz=self.tz)
            except ValueError:
                # string that cannot be parsed to Timestamp
                return ops.invalid_comparison(self, other, op)

            result = op(self.asi8, other.view('i8'))
            if isna(other):
                result.fill(nat_result)

From TDA:

        if _is_convertible_to_td(other) or other is NaT:
            try:
                other = _to_m8(other)
            except ValueError:
                # failed to parse as timedelta
                return ops.invalid_comparison(self, other, op)

            result = meth(self, other)
            if isna(other):
                result.fill(nat_result)

Note that _to_m8 means something different in these two files.

The _unbox_scalar methods might not be quite the right fit, but the idea behind them is really similar to _is_convertible_to_td and the try/except calls to to_m8.