queries: unexplained plan oscillation when running MR tpcc

[ajstorm]

When running TPC-C on the drt-large cluster, we're seeing some unexpected plan oscillation for some of the queries. We saw this earlier in the run, shortly after we loaded the tables and before the statistics were run (#122818) but then a few days later, the plans briefly oscillated again.

Discussed in this internal thread.

Jira issue: CRDB-38218

[DrewKimball]

Note: I think the current poor plan choices are happening at a more or less constant, low rate. Most executions are using the "good" plan, while certain placeholder values are using the "bad" plan. Since these tables are very large, I suspect #97701 is involved.

[mgartner]

The UPDATE on "orders" can oscilate when stats are out of date and the o_w_id value being filtered is outside the histogram range, see #64570. Also see the discussion here: https://cockroachlabs.slack.com/archives/C01RX2G8LT1/p1714421477464499

[ajstorm]

Reopening, based on a discussion with @mgartner.