inference: fix missing LimitedAccuracy handling in `conditional_cha…

…nges` (JuliaLang#40744)

The following JET analysis told me that we missed to handle
`LimitedAccuracy` in `conditional_changes` and limited conditional
constraints weren't propagated:
```julia
julia> using JET, Test

julia> report_call() do
           @testset "foo" begin; true; end
       end
┌ @ /Users/aviatesk/julia/julia/usr/share/julia/stdlib/v1.7/Test/src/Test.jl:1262 Test.finish(ts)
│┌ @ /Users/aviatesk/julia/julia/usr/share/julia/stdlib/v1.7/Test/src/Test.jl:1007 Test.print_test_results(ts)
││┌ @ /Users/aviatesk/julia/julia/usr/share/julia/stdlib/v1.7/Test/src/Test.jl:941 #self#(ts, 0)
│││┌ @ /Users/aviatesk/julia/julia/usr/share/julia/stdlib/v1.7/Test/src/Test.jl:961 Test.get_alignment(ts, 0)
││││┌ @ /Users/aviatesk/julia/julia/usr/share/julia/stdlib/v1.7/Test/src/Test.jl:1034 Test.map(JuliaLang#36, Base.getproperty(ts, :results))
│││││┌ @ abstractarray.jl:2362 Base.collect_similar(A, Base.Generator(f, A))
││││││┌ @ array.jl:638 Base._collect(cont, itr, Base.IteratorEltype(itr), Base.IteratorSize(itr))
│││││││┌ @ array.jl:723 Base.iterate(itr)
││││││││┌ @ generator.jl:47 Base.getproperty(g, :f)(Base.getindex(y, 1))
│││││││││┌ @ /Users/aviatesk/julia/julia/usr/share/julia/stdlib/v1.7/Test/src/Test.jl:1034 Test.get_alignment(t, Test.+(Core.getfield(#self#, :depth), 1))
│││││││││││││┌ @ array.jl:728 Base.collect_to_with_first!(Base._similar_for(c, Base.typeof(v1), itr, isz), v1, itr, st)
││││││││││││││┌ @ array.jl:734 Base.collect_to!(dest, itr, Base.+(i1, 1), st)
│││││││││││││││┌ @ /Users/aviatesk/julia/julia/base/array.jl:764 Base.collect_to!(new, itr, Base.+(i, 1), st)
││││││││││││││││┌ @ /Users/aviatesk/julia/julia/base/array.jl:758 Base.indexed_iterate(y, 2, _8)
│││││││││││││││││┌ @ tuple.jl:94 Base.iterate(I, state)
││││││││││││││││││ no matching method found for call signature (Tuple{typeof(iterate), Nothing, Int64}): Base.iterate(I::Nothing, state::Int64)
│││││││││││││││││└───────────────
││││││││││││││││┌ @ /Users/aviatesk/julia/julia/base/array.jl:758 Base.indexed_iterate(y, 1)
│││││││││││││││││┌ @ tuple.jl:89 Base.iterate(I)
││││││││││││││││││ no matching method found for call signature (Tuple{typeof(iterate), Nothing}): Base.iterate(I::Nothing)
│││││││││││││││││└───────────────
Union{Test.FallbackTestSet, Test.DefaultTestSet}
```

After this PR:
```julia
julia> report_call() do
           @testset "foo" begin; true; end
       end
No errors !
Union{Test.FallbackTestSet, Test.DefaultTestSet}
```

base/compiler/abstractinterpretation.jl

@@ -1846,7 +1846,12 @@ function typeinf_local(interp::AbstractInterpreter, frame::InferenceState)
 end
 
 function conditional_changes(changes::VarTable, @nospecialize(typ), var::Slot)
- if typ ⊑ (changes[slot_id(var)]::VarState).typ
+ oldtyp = (changes[slot_id(var)]::VarState).typ
+ # approximate test for `typ ∩ oldtyp` being better than `oldtyp`
+ # since we probably formed these types with `typesubstract`, the comparison is likely simple
+ if ignorelimited(typ) ⊑ ignorelimited(oldtyp)
+ # typ is better unlimited, but we may still need to compute the tmeet with the limit "causes" since we ignored those in the comparison
+ oldtyp isa LimitedAccuracy && (typ = tmerge(typ, LimitedAccuracy(Bottom, oldtyp.causes)))
  return StateUpdate(var, VarState(typ, false), changes, true)
  end
  return changes

base/compiler/typelattice.jl

@@ -91,7 +91,7 @@ struct StateUpdate
 end
 
 # Represent that the type estimate has been approximated, due to "causes"
-# (only used in abstractinterpret, doesn't appear in optimize)
+# (only used in abstract interpretion, doesn't appear in optimization)
 # N.B. in the lattice, this is epsilon smaller than `typ` (except Union{})
 struct LimitedAccuracy
  typ