Fix pint integration tests (#3600)

* don't use the index to construct the condition array

* use drop_vars to drop variables and coordinates

* update the xfail reason for astype and item

* remove the xfails due to np.result_type not being implemented

* make array_extract_units a bit more robust

* add the missing dataset apply_ufunc test

* use None as the dict key for quantities

* convert to variable to avoid the merge / assign_coords issue

relevant issue: #3483

* update the align tests

* fix the bugs introduced by converting to variables

* update a few more tests

* update the aggregation tests

* update a few more tests

* don't test rank which was deprecated in numpy 1.9

* update the DataArray.fillna tests

* update most of the remaining DataArray tests

* remove a debug assert

* fix the broadcast_equals tests

* update the indexing tests

* update the tests depending on einsum

* update the squeeze tests

* fix a bug in the head / tail / thin test design

* use dictionaries to index in the loc tests

* update the expected unit of the computation tests

* update the grouped operations tests

* update the where tests

* update most of the remaining dataset tests

* create new tests for drop_sel

* final batch of updated tests

* rename result to actual to match the other test files

* fix some more test bugs

* update the xfail marks