A more efficient implementation of the fix for MB-60046

The initial fix for MB-60046 is a minimal fix that solves the issue. However it is not the most efficient fix, although it is not proved to be performance issue.

The fix in MB-60046 means that for every auto-deleted key, there is a re-scan of the hash-chain looking for pending keys. If the hash chains are short, this may not be a big/measureable impact, but we can certainly improve.

One idea is to resurrect some draft changes from MB-36934, where the ordering of pending/committed in the hash-chain is enforced. The visitor could then always receive a pending and committed StoredValue, with out the extra chain iterations, effectively return pending and pending->next.

Note there's no evidence to suggest we should resolve this improvement, the MB-60046 fix maybe adequate