Sequential scan sampling CE is used in test suites for its determinism. However, it is highly susceptible to partially sorted data (see SERVER-123925), which can skew results.

Randomly shuffle the TPC-H datasets before running tests, then update the test suites to use the shuffled datasets. This combines the benefits of both sampling approaches: the randomness of random CE and the predictability of sequential scan sampling CE.

Ad a result, TPC-H-based tests pass consistently with shuffled input data and show no regression from the partial-sort sensitivity issue.