Taking a whole-table explicit lock might try to to scan whole table

[mbautin]

Jira Link: DB-2010
This applies to snapshot isolation. We have code in conflict_resolution.cc that tries to detect conflicts with committed transactions. If the intent we are trying to write is a lock on the whole table, we will currently try to scan all rows of the table in regular RocksDB in order to see if anything has been written after the transaction started. This is really inefficient and we should just look at hybrid time frontier values instead of that (but we'll need consensus frontiers containing committed hybrid times).

Cc @jaki @spolitov