I wouldn’t call that example minimal, pre-compiling the dependencies took more than 5 minutes, running the example another 5 minutes, and the REPL is flooded with “Interrupted. Larger maxiters is needed” warnings from the ODE solver. Ideally, a MWE should be condensed to the absolute bare minimum needed to reproduce the issue.
Terminating a second run with Ctrl+C resulted in a Julia segfault and I don’t have time today anymore to rerun the example again.
Maybe a MWE only requires a simple EnsembleProblem call. EnsembleProblem uses multithreading by default, and possibly nested multithreading causes the issues here. You could check if the issue is fixed by (a) using a different algorithm for solving the EnsembleProblem (e.g., EnsembleSerial: Parallel Ensemble Simulations · SciML) or (b) using a different parallel sampling algorithm (e.g. MCMCSerial).