Support boolean indexing with a COO sparse matrix

While investigating https://github.com/theislab/scanpy/issues/921 I ran into the following limitation when using a boolean array as an index.

This is the expected behaviour for regular NumPy:

>>> import numpy as np
x = np.array([1, 2, 3, 4])
>>> x[x % 2 == 0]
array([2, 4])

But the same thing fails for sparse:

>>> import sparse
>>> s = sparse.COO.from_numpy(x)
>>> s
<COO: shape=(4,), dtype=int64, nnz=4, fill_value=0>
>>> [s % 2 == 0]
[<COO: shape=(4,), dtype=bool, nnz=2, fill_value=True>]
>>> s[s % 2 == 0]
<class 'sparse._coo.core.COO'>
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/tom/workspace/scanpy/venv/lib/python3.7/site-packages/sparse/_coo/indexing.py", line 60, in getitem
    index = normalize_index(index, x.shape)
  File "/Users/tom/workspace/scanpy/venv/lib/python3.7/site-packages/sparse/_slicing.py", line 58, in normalize_index
    check_index(i, d)
  File "/Users/tom/workspace/scanpy/venv/lib/python3.7/site-packages/sparse/_slicing.py", line 123, in check_index
    raise IndexError("only integers, slices (`:`), ellipsis (`...`), numpy.newaxis (`None`) and "
IndexError: only integers, slices (`:`), ellipsis (`...`), numpy.newaxis (`None`) and integer or boolean arrays are valid indices

It would be good to fix it for GCXS if that's a problem too. I couldn't reproduce it there since from sparse import GCXS fails.

Start in sparse/_coo/indexing.py and sparse/_slicing.py, where the traceback shows boolean index normalization failing for a COO array. Trace how boolean arrays are validated and indexed, then check whether the same behavior applies to GCXS. Done means the shown boolean-indexing example works for COO and, if applicable, GCXS without breaking existing index forms.