feat: refine .split() return type

[castarco]

NOTE: ⚠️ Still work in progress. ⚠️

PR Description

This PR refines String.split(*) return type (plus the RegExp's cousin method).

As opposed to previous attempts, it does not try to characterise the "empty array result" as a special case, but quite the opposite, it characterises "[string, ...string[]] result" as a special case.

With this change, we only get [string, ...string[]] as return type when we can statically know for sure that we are not trying to execute ''.split(''), and we do not set the second parameter (limit) to 0.

In other words, the separator must be a literal for this to give as a "nice" return type. If we use a variable, we'll keep the more general string[] return type.

This (combined with the fact that we can't specialise interfaces for specific literals) means that there are cases where we still get non-empty arrays as a result but our return type won't be able to represent that fact.

Other Relevant Details

I introduced the types NonEmptyStringParam<> and NonZeroNumberParam<>. I suspect that these 2 new types might be not well received. I'm not sure how to make the non-global.

Pending tasks:

• Fix formatting issues
• Fix baseline tests
• Add new tests

What this PR tries to fix

• Fixes Improve string split return type of first array index #53362

[MartinJohns]

No idea why you decided to open another PR instead of just changing your existing. Your PR is still about the same issue, and splitting it like this just scatters the discussion.

Nothing proposed here hasn't been proposed before. See also: #53362 (comment)

we have a pretty low appetite for making the standard library definitions extremely complex for the sake of NUIA

I'd argue your suggested change is very complex. See also: #56333 (comment)

[castarco]

@MartinJohns Ok. 🤷🏻 . I guess there's no point on fixing the tests then. (and sorry for the duplicate PR).