[Dimensionality reduction] Add metrics based on NN co-ranking

[michalk8]

What is the metric?
Describe it briefly and include a citation if applicable.
There are many metrics described here: https://core.ac.uk/download/pdf/148668147.pdf

How should it be implemented?
Include a link to the publicly available codebase if available.
https://www.sciencedirect.com/science/article/pii/S2405844021003042
contains the code. Scaling will also be an issue (for more than 3k cells [estimate]).
I'd also double-check if the implementation is correct.

Which task(s) could it be used for?
If the method is to be used for a task that is not yet included in the code base, use the issue template propose a new task instead.
Dimensionality reduction (2D), possibly

[bendemeo]

This is a great reference! I think co-k-nearest-neighbor size (Q_NN(k)) is especially useful, although it has a nuisance parameter k. To make it more robust, we might average Q_NN(k) for k up to some threshold, possibly weighting for lower distances. But maybe Q_NN(k) is a good place to start. Not sure if the original authors compute an entire distance matrix, but I could write code to generate a sparse Q_NN matrix using only nearest neighbor data from sklearn.neighbors.