The initial implementation for SQL Stats flush logic performs as follow:

1. It tries to run an INSERT ... ON CONFLICT DO NOTHING statement for each stats object.
2. If it encounters an conflict, it reads the statistics column from the conflicting row and update it with new stats using SELECT ... FOR UPDATE statement
3. It runs a UPDATE statement to update the rows.

Note that for the fast path, we are performing 1 network round trip, and for slow path, we are performing 3 network round trip.

We can amortize this by batch multiple stats object into a single requests and improve the throughput.

Jira issue: CRDB-8749