Add sparsificiation step before sparse-dependent Scrublet calls

[pinin4fjords]

This PR addresses #1645, which was caused by my injection of normalised matrices from Scanpy workflows, thereby bypassing a sparseness check Scrublet does with the raw matrix.

The fix contained here is apply the Scrublet sparseness check, before calling Scrublet functions with a dependency on sparseness.

[ivirshup]

Thanks for the fix! And test!

---

Minor point on import, typically scipy.sparse isn't a way you can access that attribute:

In [1]: import scipy

In [2]: scipy.sparse
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-2-f25716665fba> in <module>
----> 1 scipy.sparse

AttributeError: module 'scipy' has no attribute 'sparse'

I think it works here because of the order of imports (anndata import scipy.sparse, so it becomes visible), but that may change in the future. Could you import the sparse module directly instead?

from scipy import sparse

[ivirshup]

Great, thanks!

[ivirshup]

@meeseeksdev backport to 1.7.x