Update performance test structure

[katxiao]

Problem Description

Performance tests can become more generalized if we specify the expected runtime per dataset, instead of per individual test cases.

Expected behavior

The current performance tests are based on test cases that specify the transformer, the dataset generator and the sizes.

This should be changed to a more generic approach where each dataset generator indicates its expected output values for different data sizes, and the performance test simply takes a transformer and runs all the compatible datasets through it, validating against those expected output values.