Validate inputs and outputs earlier.

[maleadt]

Metal's LLVM post-processing inspects arguments (as does CUDA.jl's), so we need to make sure there's no invalid stuff in there. Currently, we were only running the code instance validation during emit_asm, instead make it happen as early as possible, during emit_julia.

  Got exception outside of a @test
  LoadError: Type Array does not have a definite size.
  Stacktrace:
    [1] sizeof(x::Type)
      @ Base ./essentials.jl:473
    [2] add_argument_metadata!(job::GPUCompiler.CompilerJob, mod::LLVM.Module, entry::LLVM.Function)
      @ GPUCompiler ~/Julia/pkg/GPUCompiler/src/metal.jl:612
    [3] finish_ir!(job::GPUCompiler.CompilerJob{GPUCompiler.MetalCompilerTarget}, mod::LLVM.Module, entry::LLVM.Function)
      @ GPUCompiler ~/Julia/pkg/GPUCompiler/src/metal.jl:108
    [4] macro expansion
      @ ~/Julia/pkg/GPUCompiler/src/driver.jl:400 [inlined]
    [5] macro expansion
      @ ~/Julia/depot/packages/TimerOutputs/LHjFw/src/TimerOutput.jl:253 [inlined]
    [6] macro expansion
      @ ~/Julia/pkg/GPUCompiler/src/driver.jl:340 [inlined]
    [7] emit_llvm(job::GPUCompiler.CompilerJob, method_instance::Any; libraries::Bool, deferred_codegen::Bool, optimize::Bool, cleanup::Bool, only_entry::Bool, ctx::LLVM.Context)
      @ GPUCompiler ~/Julia/pkg/GPUCompiler/src/utils.jl:83
    [8] mtlfunction_compile(job::GPUCompiler.CompilerJob, ctx::LLVM.Context)
      @ Metal ~/Julia/pkg/Metal/src/compiler/execution.jl:166

[maleadt]

Enzyme.jl failure unrelated.