cache Type{wrapper} for faster lookups

[JeffBezanson]

Benchmark from #44402:

julia> @btime Core.Compiler.widenconst(c1)
  98.999 ns (0 allocations: 0 bytes)
Type{Base.RefValue}

julia> @btime Core.Compiler.widenconst(c2)
  25.269 ns (0 allocations: 0 bytes)

alternative to #44447

mostly fixes #44402

[aviatesk]

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

[nanosoldier]

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

[vtjnash]

Suggested change

	jl_serialize_value(s, tn->Typeofwrapper);
	jl_serialize_value_(s, (jl_value_t*)tn->cache, 0);
	jl_serialize_value_(s, (jl_value_t*)tn->linearcache, 0);
	jl_serialize_value(s, jl_atomic_load_acquire(&tn->Typeofwrapper));
	jl_serialize_value_(s, (jl_value_t*)jl_atomic_load_acquire(&tn->cache), 0);
	jl_serialize_value_(s, (jl_value_t*)jl_atomic_load_acquire(&tn->linearcache), 0);

(I don't yet have this file enabled for this static-analysis check, since there are other similar updates missing too)