nospecialize a few more things that gain many specializations without having significant runtime

[KristofferC]

I used SnoopCompile (JuliaDebug/SnoopCompile.jl#462 needed) to basically run https://juliadebug.github.io/SnoopCompile.jl/dev/tutorials/pgdsgui/ on creating an ODEProblem and looking at methods that had many specializations

  Top 30 methods by inclusive inference time
==================================================================================================================================
Rank  Method                                    Module                Inclusive(s)  Exclusive(s)  #Specs  Location
----------------------------------------------------------------------------------------------------------------------------------
   1  #get_A_b_from_LinearFunction#918          ModelingToolkitBase         1.8858        1.8352      89  /Users/kc/JuliaTests/SymbolicUtilPerf/ModelingToolkit.jl/lib/ModelingToolkitBase/src/problems/linearproblem.jl:125
   2  process_SciMLProblem                      ModelingToolkitBase         1.6369        0.0560      14  /Users/kc/JuliaTests/SymbolicUtilPerf/ModelingToolkit.jl/lib/ModelingToolkitBase/src/systems/problem_utils.jl:1511
   3  #_#795                                    ModelingToolkitBase         1.5327        0.0075       1  /Users/kc/JuliaTests/SymbolicUtilPerf/ModelingToolkit.jl/lib/ModelingToolkitBase/src/problems/odeproblem.jl:101
   4  ODEProblem                                ModelingToolkitBase         1.5317        0.0023       1  none:0
   5  #ODEProblem#793                           ModelingToolkitBase         1.5293        0.0008       2  none:0
   6  #_#794                                    ModelingToolkitBase         1.5276        0.0008       3  none:0
   7  ODEProblem                                ModelingToolkitBase         1.5273        0.0004       3  none:0
   8  ODEProblem                                ModelingToolkitBase         1.5264        0.0005       3  /Users/kc/JuliaTests/SymbolicUtilPerf/ModelingToolkit.jl/lib/ModelingToolkitBase/src/problems/odeproblem.jl:101
   9  #build_explicit_observed_function#699     ModelingToolkitBase         0.9100        0.5522      30  /Users/kc/JuliaTests/SymbolicUtilPerf/ModelingToolkit.jl/lib/ModelingToolkitBase/src/systems/codegen.jl:1219

This m method despecializes some of these and timing compile time and runtime for

using Multibody, ModelingToolkit

@named model = Multibody.Robot6DOF();
ssys = multibody(model)
prob = ODEProblem(ssys, [
                   ssys.mechanics.r1.phi => deg2rad(-60)
                   ssys.mechanics.r2.phi => deg2rad(20)
                   ssys.mechanics.r3.phi => deg2rad(90)
                   ssys.mechanics.r4.phi => deg2rad(0)
                   ssys.mechanics.r5.phi => deg2rad(-110)
                   ssys.mechanics.r6.phi => deg2rad(0)

                   ssys.axis1.motor.Jmotor.phi => deg2rad(-60) * (-105) # Multiply by gear ratio
                   ssys.axis2.motor.Jmotor.phi => deg2rad(20) * (210)
                   ssys.axis3.motor.Jmotor.phi => deg2rad(90) * (60)

                   ssys.axis1.motor.Jmotor.w => 0
                   ssys.axis2.motor.Jmotor.w => 0
                   ssys.axis3.motor.Jmotor.w => 0
               ], (0.0, 2.0));

I get for compile time and runtime respectively:

# PR:
# 20.132280 seconds (154.21 M allocations: 8.043 GiB, 7.63% gc time, 67.38% compilation time: 17% of which was recompilation)
# 5.196923 seconds (64.45 M allocations: 3.164 GiB, 14.40% gc time, 1.38% compilation time)

# Master:
#  23.805818 seconds (169.28 M allocations: 8.799 GiB, 8.30% gc time, 71.23% compilation time: 13% of which was recompilation)
#   5.201396 seconds (64.71 M allocations: 3.185 GiB, 15.22% gc time, 2.45% compilation time)

so no observed reduction in runtime but around 15% reduced compile time.