Add `remove_pseudocount` option in get_V_row and get_V_column methods

[scjlee]

Previously I manually remove pseudocounts by manually putting modified V_row and V_column into the scaling. However, the current implementation prevent this work around.

We probably need to implement options letting user can control whether to remove pseudocounts during calculation.

[denklewer]

Shouldn't we deal with pseudocounts before passing to the method? Won't this brake the whole maths there if we modify V_row V_column?

[scjlee]

But remember that if data is too sparse, sinkhorn might not converge or would took very long time to converge. In single-cell data it happened.

[denklewer]

But you don't you do it before sinkhorn?

[scjlee]

Adding pseudocount is to facilitate convergence of Sinkhorn). Precisely, I remove pseudocount during projection.

I feel I don't provide rigorous proof though this approach seems to work. I will do it outside the object for now. This probably make more sense.