respecialize normalized vararg type parameters as part of precise_container_type

[jrevels]

Fix #22875:

julia> @inline g(args...) = one(args...)
g (generic function with 1 method)

julia> f(args...) = g(args...)
f (generic function with 1 method)

# this previously inferred as Any
julia> @code_warntype f(Int)
Variables:
  #self#::#f
  args::Any

Body:
  begin
      return 1
  end::Int64

# this was also previously inferred as Any
julia> @code_warntype Base.promote_type(Int, Float64, Int, Int, Int, Int)
Variables:
  #self#::Base.#promote_type
  T::Any
  S::Any
  U::Any
  V::Any

Body:
  begin
      return Float64
  end::Type{Float64}

This still needs tests, which I'll get to tomorrow (or after dinner, if I'm feeling plucky).

Many thanks to @vtjnash for once again guiding me through the dark waters of inference.

[jrevels]

Travis failure unrelated?

[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

[jrevels]

Can't reproduce the regressions locally. I can reproduce the perf_local_array speedups, though. It seems that before this PR, that benchmark was hitting #22875 via the calls to samerand(T, n), causing downstream variables to be inferred as Any. With this fix, the function can be fully inferred.