Allow NaN values in unsupervised random forest

[weioren]

trafficstars

Reference Issues/PRs

What does this implement/fix? Explain your changes.

Changed fit function in _unsupervised_forest.py to add ensure_all_finite = "allow-nan" parameter to ensure NaN values are always allowed.

Any other comments?

[adam2392]

Hi @weioren note this will not enable splitting on missing values, as the core logic within Cython needs to get implemented.