Fix to p-value calculation for permutation and exact

[mjacketti]

The discrepancy was caused by a tail-direction error in the exact and permutation p-value calculations. The JT statistic increases as the evidence for an increasing trend gets stronger. Therefore, the increasing alternative should use the upper tail of the null distribution, P(J ≥ observed J), while the decreasing alternative should use the lower tail, P(J ≤ observed J). The code was effectively using the opposite tail for the permutation method, and the exact method was similarly misaligned with the statistic direction. As a result, datasets with a very strong increasing trend produced p-values near 1 instead of near 0. The approximate method used the expected direction, which is why it produced the statistically sensible result.

[adavis10]

Everything looks good. I think merging into main is fine b/c we're not changing any of the documentation so the monthly scans won't pull anything not ready for public release into the scheduled updates.