javalib Files#walk FileSystemLoop detection is probably wrong/broken.

[LeeTibbert]

Whilst studying Issue #3744, I have come to suspect that the algorithm in Files#walk to detect file
system cycles and throw FileSystemLoopException can report both false positives and, worse,
false negatives (i.e. loop without termination).

Files#find uses walk(), so it is suspect also.

I think I must plead guilty on this one, even in a community with a ego-less programming norm.

I am currently working on File.walkFileTree. If my suspicions about Files#walk tree bear out
I hope to fold the loop detection algorithm from there into its sibling Files#walk.

This is going to take a long while, so no problem if somebody gets there before me.
Hint: consider back-edges visited, where a back-edge can be a directory ancestor
of a link used to get to the current path or such a link itself.

[LeeTibbert]

merged PR #3917 fixed a series of bugs with symlink cycle detection (Thank you Wojciech!).
In particular, it solved my concerns with false positives where a symlink referred to a previously visited regular file.
I think it also fixed my concern with false negatives.

I will open another Issue detailing two specific defects which still exist. Rome was not built in a day.