support for sparse matrices in `linalg.clarkson_woodruff_transform`

```pycon
>>> import numpy as np; import scipy
>>> a = scipy.sparse.coo_array([[1,2,3], [2,3,4], [4,5,6]])
>>> scipy.linalg.clarkson_woodruff_transform(a, 2)
<Compressed Sparse Column sparse matrix of dtype 'int64'
	with 3 stored elements and shape (2, 3)>
>>> scipy.linalg.clarkson_woodruff_transform(a.toarray(), 2)
array([[4, 5, 6],
       [3, 5, 7]])
```