Performance regression: iterating Tuples of Functions allocates

[MarcMush]

using BenchmarkTools
op = (+,-,*,/,\)
f() = [1 for _=op, _=op]
@btime f()

1.10.6: 575.956 ns (1 allocation: 256 bytes)
1.11.1: 11.000 μs (119 allocations: 3.23 KiB)
1.12.0-DEV.1640: 9.900 μs (119 allocations: 3.23 KiB)

in my testing,

• op must have at least 5 different functions
• op has to be iterated at least twice
• op must be a tuple (not Array)
• op being global or const has no impact

[MarcMush]

Bisected to a61d1b4 (#50927)

a61d1b47a68297704814188a3509c011ec2a8fa1 is the first bad commit
commit a61d1b47a68297704814188a3509c011ec2a8fa1
Author: Jameson Nash <vtjnash@gmail.com>
Date:   Fri Sep 15 18:32:16 2023 -0400

    inference: apply tmerge limit elementwise to the Union (#50927)

    This allows forming larger unions, as long as each element in the Union
    is both relatively distinct and relatively simple. For example:

        tmerge(Base.BitSigned, Nothing) == Union{Nothing, Int128, Int16, Int32, Int64, Int8}
        tmerge(Tuple{Base.BitSigned, Int}, Nothing) == Union{Nothing, Tuple{Any, Int64}}
        tmerge(AbstractVector{Int}, Vector) == AbstractVector

    Disables a test from dc8d885, which does not seem possible to handle currently.

    This makes somewhat drastic changes to make this algorithm more
    commutative and simpler, since we dropped the final widening to `Any`.

    Co-authored-by: pchintalapudi <34727397+pchintalapudi@users.noreply.github.com>
    Co-authored-by: Oscar Smith <oscardssmith@gmail.com>

[oscardssmith]

that's somewhat surprising. That PR usually makes inference stronger.