Some test improvements I noticed while working on https://github.com/10gen/mongo/pull/60458

Replace sleeps + timing assumptions in oplog_truncation_size_based.js with failpoints and retries. This is going to be more important after SERVER-132836 once we're always sampling to create the initial markers, because sampling is faster than scanning and initial truncation happens sooner.

• Use hangOplogCapMaintainerThread failpoint to hold off truncation until after initial checks are done
• Retry on CappedPositionLost when running queries against the oplog collection
• Use assert.soon retries on more checks