Simplify zip take 2

[haampie]

This pr:

1. Removes Zip2 since Zip1 is a valid base case
2. Fixes and adds a test for an isdone bug for the Zip iterator.
3. Flattens the state like the values: iterate(zip(a, b, c)) returns ((val_a, val_b, val_c), (state_a, state_b, state_c)).
4. Removes a call to isdone(z.b) in the iterate function, since that is handled by recursion already when calling iterate(z.b, ...).
5. Adds a macro that @iamed2 suggested to "bail on nothing"; there's this three-liner pattern y = produce(); y === nothing && return nothing; a, b = y all over the place, which is a, b = @something produce() with the macro. Yet I suspect people wouldn't like this macro in base, at least not with this name. Suggestions?
6. Adds a note to the docs about stateful iterators & how they are not advanced when another iterator finishes in the current iteration.

[haampie]

Apparently the DRY-fix was breaking inference, even on the current master.

This PR with the last commit:

> @code_warntype iterate(zip(1:2,1:2,1:2,1:2), (1,1,1,1))
Body::Union{Nothing, Tuple{NTuple{4,Int64},NTuple{4,Int64}}}

Current master of Julia:

> @code_warntype iterate(zip(1:2,1:2,1:2,1:2), (1,(1,(1,1))))
Body::Union{Nothing, Tuple{Tuple{Int64,Any,Vararg{Any,N} where N},Tuple{Int64,Tuple{Any,Union{Tuple{Any,Any}, Base.LegacyIterationCompat{_1,_2,_3} where _3 where _2 where _1}}}}}

I don't yet see an easy way to test inference for Union{Nothing, ...} types, so a test is yet to be added.

[Keno]

I don't yet see an easy way to test inference for Union{Nothing, ...} types, so a test is yet to be added.

You can use Compiler.typesubtract on the inferred type.

[haampie]

I've added the commit of #27516 with an inference test of the new implementation of zip. Hopefully this can be merged soon :)

[KristofferC]

@nanosoldier runbenchmarks(ALL, vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

[haampie]

Superseded by #29238