To avoid the need to mess with `lit` tests due to renamed benchmark, let's just add the customary suffix `2` for the bugfixed version. We'll fight the renaming battle another day...
Remove the extended benchmark coverage to leave only bug-fixed version of `RomanNumeral` benchmark with the new name `Roman.Substring.startsWith.dropFirst`.
Rewrite of the `RomanNumbers` benchmark that fixes error in the `position` which was always reset to 0 after `continue outer`, by swapping the `while` and `for-in` loops into more natural order, which eliminates the need for using `position` at all. This is based on examples from [Rosetta Code](https://www.rosettacode.org/wiki/Roman_numerals/Decode#Swift).
The coverage has been expanded to measure performance of `String`/`Substring`/`UTF8View` with very short string arguments and methods: `hasPrefix`/`starts(with:)`, `removeFirst`/`dropFirst` and `String.append()`, as well as functional variants of the fastest imperative version.
The `Roman.Substring.startsWith.dropFirst` variant tests the same code paths as the original `RomanNumbers` benchmark, but the name is changed because of the bug-fix and lowered workload (to run in less than 1000 μs).
For comparison, there's one extra variant with character based parsing algorithm: `Roman.DictCharInt.map.reduce`.
