Removes the restriction on the amount of non-void Parsers that can be used in a Parserbuilder

[coenttb]

This pull request eliminates the need for all Take*n overloads except for Take2 while maintaining full backward compatibility.

The Take2.Map parameter pack construct is a deliberate tradeoff. Ideally, we would prefer to use a Conversion with a MapConversion for a more elegant solution. However, consider the following hypothetical TupleConversion:

public struct TupleConversion<each T1, T2>: Conversion {
    public typealias Input = ((repeat each T1), T2)
    public typealias Output = (repeat each T1, T2)

    public func apply(input: ((repeat each T1), T2)) -> (repeat each T1, T2) {
        let (first, second) = input
        return (repeat each first, second)
    }

    public func unapply(output: (repeat each T1, T2)) -> ((repeat each T1), T2) {
        guard let tuple = output as? ((repeat each T1), T2) else {
            fatalError("Could not convert output tuple type")
        }
        return tuple
    }
}

While functional, this approach depends on macOS 14 for its use of parameter packs in generics, which makes it incompatible with earlier versions. As a result, introducing Take2.Map strikes a practical balance: it ensures backward compatibility while adding only minimal complexity to the surface area.

For verification, refer to VariadicTests.swift, which includes passing tests.

PS: I recommend exploring further if and to what extent the existing Map infrastructure can be leveraged over Take2.Map.

[stephencelis]

Thanks for looking into this! I'll bump the package file on main after merging since parameter packs require 5.9, but that's been around for awhile now.

Feel free to experiment with your TupleParser using compiler conditionals or availability checking in another PR if you think it'll improve performance on platforms that support it!