relax maxOpenTransactions restriction for recovery tools

…on closure

Race between metatable generation and transaction closure

…conversions

…le-casting

Fixed casting issue: skipped Index Condition Pushdown optm for lossy conversions

getCostRangeIndex() was giving an index that only matches the leading
column(s) of an ORDER BY nearly the same sort discount as an index that
matches the whole order. A real column mismatch stopped the loop early
but didn't disqualify the coveringCount credit already picked up before
the mismatch, so two very different indexes could tie in cost even
though one still needs a real sort within each equality group and the
other doesn't.

Only disqualify the discount when the index has a different column at
that position. A direction/null-ordering mismatch on the same column is
left alone, since that's still usable via a reverse index scan and the
existing indexes.sql tests rely on the partial credit there.

Fixes #4353

…-sort-match

Fix getCostRangeIndex() giving partial-match indexes an unfair sort discount