[benchmark] Add to docs benchmark on how fast each transform is

[ternaus]

Currently, the CPU is bottlenecked, and it is unclear which augmentations will give the biggest speedup.

=> We need code for the benchmark and a page in the docs that shows the speed of each transform.

Some algorithms' performance depends on input types and parameters. For example, some algorithms have fast / approximate and exact implementations.

This could also be taken into account.