Regression (change) in inference after SCC PR

[KristofferC]

#39116 caused the following change:

julia> using MutableArithmetics

julia> MutableArithmetics.copy_if_mutable(Vector{Int}[])
Any[]
# Vector{Int64}[] before #39116

What is interesting is that when running it through the interpreter it still gives the older ("better") result

julia> using JuliaInterpreter

julia> @interpret MutableArithmetics.copy_if_mutable(Vector{Int}[])
Vector{Int64}[]

This caused a few packages to start failing in the backport branch (for example https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/7d1533d_vs_8240225/MultivariateBases.1.6.0-beta1-85c46c0116.log).

I'll put a 1.6 blocker on this for now, but I don't know if this is expected or how it should be assessed.

cc @vtjnash