Trivial optimization in generate_initializesystem

[hersle]

During profiling of ODEProblem in this example of my package, I found I saw being slowed down a lot by the line that calls default_toterm on every element in keys(defs) for every iteration in the loop, which I think could simply be precomputed before the loop.

using SymBoltz # https://github.com/hersle/SymBoltz.jl
using BenchmarkTools
M = ΛCDM() # a big System
Ms = mtkcompile(M)
pars = SymBoltz.parameters_Planck18(M) # some sym-to-num parameter dict
pars[M.k] = 1.0
@btime prob = ODEProblem(Ms, pars, (0.0, 10.0))

Before PR:

  2.607 s (27144415 allocations: 803.71 MiB)

After PR:

  995.616 ms (9386044 allocations: 317.20 MiB)

These changes give me a 3x speedup in construction of a big problem, and I think the logic should be equivalent before and after.

[hersle]

One more

  717.067 ms (8363734 allocations: 280.19 MiB)

[AayushSabharwal]

Awesome, thanks for noticing this!

[AayushSabharwal]

Waiting on SciML/NonlinearSolve.jl#634 to rerun CI

[hersle]

Re-run now and, if problem-free, get this in before the next release? 😃 @AayushSabharwal