Compute table candidates with last buffer current

[clemera]

As discussed in #75 and #312 describe-variable was slowed down with Selectrum because the predicate switches back to the last buffer for each candidate. By computing the candidates with the last buffer current by default we avoid the slow down for such cases. The docs don't mention any assumptions about the current buffer for the table/predicate, Ivy does run the computation before the minibuffer is even entered so it doesn't has this problem, we keep computing them late to allow for example selectrum-set-selected-candidate to skip the initial computation which is used by embark for example.