:random mode in find-plans is quite inefficient

[rpgoldman]

Currently it is implemented by doing a non-destructive random permutation of all options.

In many cases, though, we do not need to backtrack -- a random set of planning decisions should always yield a plan, or at least almost always do so. In that case it would be better to randomly choose one alternative. Possibly this could be implemented using a :random-one modification of :random mode.