improve inferrability of create_expr_cache

[ranocha]

I improved type stability to fix some invalidations. This is based on the following code:

julia> import Pkg; Pkg.activate(temp=true); Pkg.add("HDF5")

julia> using SnoopCompileCore; invalidations = @snoopr(using HDF5); using SnoopCompile

julia> trees = invalidation_trees(invalidations)
...
 inserting write(obj::HDF5.Dataset, x) in HDF5 at ~.julia/packages/HDF5/T4H0V/src/datasets.jl:296 invalidated:
   mt_backedges: 1: signature Tuple{typeof(write), Any, String} triggered MethodInstance for Base.create_expr_cache(::Base.PkgId, ::String, ::String, ::Vector{Pair{Base.PkgId, UInt64}}, ::IOBuffer, ::Base.DevNull) (1 children)
                 2: signature Tuple{typeof(write), Any, String} triggered MethodInstance for Base.create_expr_cache(::Base.PkgId, ::String, ::String, ::Vector{Pair{Base.PkgId, UInt64}}, ::IO, ::IO) (21 children)
   24 mt_cache

The majority of invalidations from loading HDF5.jl are fixed by JuliaLang/Pkg.jl#3202 but I discovered these along the way, too, so I can also fix them here.

[ranocha]

Bump

[ranocha]

Bump

[ranocha]

Bump

[KristofferC]

CI looks really sad so I rebased it again.

[PallHaraldsson]

The has already been reviewed/"approved", just Jeff busy or waiting for CI to not fail or just end...?

PowerPC is still running for a long time, has failures (false alarms?) and/or CI needs restarting?

[KristofferC]

I think the CI failure in here is related so they have to be fixed for this to be able to get merged.

[ranocha]

CI runs of buildkite fail with

Error in testset compiler/codegen:
  | Error During Test at /cache/build/default-amdci4-1/julialang/julia-master/julia-5741f5740d/share/julia/test/compiler/codegen.jl:698
  | Test threw exception
  | Expression: readchomp(`$pfx/bin/$(Base.julia_exename()) -e 'print("no codegen!\n")'`) == "no codegen!"
  | failed process: Process(`/tmp/jl_KDMiFk/bin/julia -e 'print("no codegen!\n")'`, ProcessExited(1)) [1]

Does anybody know how this might be related?