Block bootstrap for SH-test (etc)

[seb-mueller]

Having done some test with the SH.test function, I've noticed that comparing trees constructed from adjacent blocks of a multiple alignment often reject H0 (trees constructed by adjacent blocks are different).
I suspect this is because could be due to local correlation.

The implemented bootstrap approach only operates on drawing with replacing which doesn't preserve the order:

phangorn/R/SH.R

Line 77 in b183fd6

boot <- tabulate(sample(wvec, replace=TRUE), nbins=lw)

I was therefore wondering if the block bootstrap might be more appropriate for this scenario (or possibly in general since all sequences might have local correlation.
This has also been discussed in Holmes 2012 to a degree.

I was therefore wondering if the block bootstrap makes sense and given it does whether it could be implemented?