Type-annotated destructuring assignment can crash

[mlechu]

Discussed a bit at #60733 (comment)

julia> (q,r,s)=1,1,1 # just to prevent UndefVarError, though I'm not sure that should happen either
(1, 1, 1)

julia> (q,r,s)::Tuple = (1,2,3)
(1, 2, 3)

julia> ((q,r),s)::Tuple = ((1,2),3)
[65584] signal 11 (2): Segmentation fault: 11

julia> Meta.@lower ((q,r),s)::Tuple = ((1,2),3)
:($(Expr(:thunk, CodeInfo(
1 ─ %1  = Main.q
│   %2  = Main.r
│           builtin Core.tuple(%1, %2)
│   %4  = Main.s
│   %5  =   builtin Core.tuple(%8, %4)            # <- can't access %8
│   %6  = Main.Tuple
│           builtin Core.typeassert(%5, %6)
│   %8  =   builtin Core.tuple(1, 2)
│           builtin Core.declare_global(Main, :s, true)
│         $(Expr(:latestworld))
│   %11 =   builtin Core.get_binding_type(Main, :s)
│         #s3 = 3
│   %13 = #s3
│   %14 =   builtin %13 isa %11
└──       goto #3 if not %14
2 ─       goto #4
3 ─ %17 = #s3
└──       #s3 = Base.convert(%11, %17)
4 ┄ %19 = #s3
│           builtin Core.setglobal!(Main, :s, %19)
│   %21 =   builtin Core.tuple(%8, 3)
└──       return %21
))))

Found while writing the AST validator; not sure if I should consider it valid syntax (and fix lowering) or consider it invalid (and protect lowering from this) given how many versions of julia it fails on