Lower priority for speculative variadic tuple inferences

[ahejlsberg]

This PR reverts #39281 and #39614 and instead implements the originally envisioned solution of giving higher priority to inferences made to [...T, X] than inferences made to [...T, X?] (because the latter speculates that the optional element is present). For example:

declare function ft<T extends unknown[]>(t1: [...T], t2: [...T, number?]): T;

ft([1, 2, 3], [1, 2, 3]);  // [1, 2, 3]
ft([1, 2], [1, 2, 3]);  // [1, 2]
ft(['a', 'b'], ['c', 'd']);  // ['a', 'b']
ft(['a', 'b'], ['c', 'd', 42]);  // ['a', 'b']

Fixes #39253.

[DanielRosenwasser]

I think without the explanation, this is unclear.

[DanielRosenwasser]

If the documentation is going to be sparse on SpeculativeTuple, I think showing a concrete example here is worthwhile. Your explanation on the PR would be great.