performance improvements for inverse functions of signrank and wilcox

[ArnoStrouwen]

Currently, these functions perform a linear search over the entire range.
More advanced search methods, like binary search, would greatly improve performance.
There is also symmetry which could be exploited.

This has not yet been done because getting the behavior right is quite subtle and is the cause of bugs, e.g. for binomial:
#170
Given that these functions see very little practical use, it is not a priority.

[ArnoStrouwen]

Furthermore, the DP array could be reused between different cdf calls.