How to avoid `Did not converge after `maxiters = 1000` substitutions.` warnings?

[TorkelE]

We currently get a very large* number of these in the Catalyst tests, clogging up the test log. Here is a small MTK example which generates them:

# Declares the model (with vector species/parameters, with/without default values, and observables).
using ModelingToolkit
using ModelingToolkit: t_nounits as t, D_nounits as D

# Creates the model.
@variables (X(t))[1:2]
@parameters p[1:2]
eqs = [
    0 ~ p[1] - X[1],
    0 ~ p[2] - X[2]
]
@named nlsys = NonlinearSystem(eqs)
nlsys = complete(nlsys)

# Creates the `NonlinearProblem`.
u0 = [X => [1.0, 2.0]]
ps = [p => [4.0, 5.0]]
base_nlprob = NonlinearProblem(nlsys, u0, ps)

┌ Warning: Did not converge after `maxiters = 1000` substitutions. Either there is a cycle in the rules or `maxiters` needs to be higher.
└ @ Symbolics C:\Users\Torkel\.julia\packages\Symbolics\P3CEu\src\variable.jl:574