test speedup for slowest/pathological tests (round 3)

More fixing of tests, following the same formula as LUCENE-9163.

I tried to focus again on tests that can randomly "explode", applied a few more techniques to sort them out, for example running all tests with SimpleText codec and suppressing that codec for the ones that really struggle with it, running with different Directory implementations, etc.

These kinds of outliers will randomly send test times sky high: so while this patch may only save a minute or so for the best case, I think it really helps reduce the stddev and worst case.