Improve benchmarks

[Uzlopak]

We should first make some good benchmarks for strings, with various cases. E.g. I think one of the tests is repeated NUL-character (\u0000), which is basically pointless, because it should have the same performance like a string with length of 1 with one NUL character. So we should make some smart test cases and then balance the performance.

Also using tinybench, as it is more correct than any other benchmark engine right now on the market. E.g. one of the tables in benchmark.js is mathematically wrong. mitata is programmatically not correct.

Originally posted by @Uzlopak in #792 (comment)