`@compile_workload` does not include compilation for macro expansion of body

[topolarity]

Consider a package like:

module DroppedPrecompile

using PrecompileTools
macro fancy_macro()
    f = Core.compilerbarrier(:type, length)
    return f(Tuple{Int64,Int32,Int16,Int8}[]) # this does not end up pre-compiled
end

@compile_workload begin
    @fancy_macro
end

end # module DroppedPrecompile

The dynamically called Base.length(::Tuple{Int64,Int32,Int16,Int8}) does not end up pre-compiled:

julia> using DroppedPrecompile
julia> Base.method_instance(Base.length, (Vector{Tuple{Int64,Int32,Int16,Int8}},)).cache
ERROR: UndefRefError: access to undefined reference

Logically, this makes sense - the macro expansion happens before the begin ... end block actually starts to execute.

Nonetheless, I wanted to file an issue, since it is a practical issue for libraries like ModelingToolkit.jl, which hit this use case (https://github.com/SciML/ModelingToolkit.jl/blob/3f965be3d256412844f7d43262d75fb48657c2df/src/ModelingToolkit.jl#L245-L278) and probably do intend to pre-compile the macros themselves, so that these are fast for future users.

Is there any good option for MTK to "opt-in" to including macro expansion in the workload?

[topolarity]

cc @ChrisRackauckas