Implements out keyword for where and reductions and order keyword for where (#1643)

* Implements `out` keyword into reductions

* Implements `order` keyword into `dpt.where`

* Implements `out` kwarg in `where`

* Adds new tests for order kwarg in where to test_where_order

* Adds test_where_out

* Adds test_where_out_arg_validation

* Permit overlap of inputs and output to `where` when tensors are logically equivalent

* Tweak to sum docstring

* Rearranges _comparison_over_axis and _search_over_axis to remove fast paths for zero-size arrays

This prevents possible edge cases where an array with a non-zero number of elements could be allocated prior to an error being thrown for a reduction being performed over a size-zero axis

* Adds tests for reduction out kwarg

* Applies suggestions per PR review

Avoids a call to permute_dims in reductions when `axis=None`

Make reduction tests more efficient by reusing an array of zeros rather than reconstructing

Author: ndgrigorian