FastBroadcast.jl is not using multithreading with `ArrayPartition`

Things like this are not multithreaded when u is an ArrayPartition:

@.. broadcast=false thread=true u=u + x * u

ArrayPartitions use a custom BroadcastStyle. FastBroadcast.jl therefore doesn't go into the fast_materialize_threaded! branch and instead calls the generic materialize!, which is not threaded:
https://github.com/YingboMa/FastBroadcast.jl/blob/ad586d83ffcac15c92969b93dd5cf0c8fd025af9/src/FastBroadcast.jl#L305-L313

Multithreaded schemes in OrdinaryDiffEq.jl are using @.. broadcast=false thread=thread ..., and OrdinaryDiffEq.jl is using an ArrayPartition for DynamicalODEProblems. Therefore, time integration is not multithreaded (even when setting thread=True()) with a DynamicalODEProblem and a scheme that works for a general ODEProblem (including all RK methods).

Not sure in which repo this should be fixed, so I just reported it here.

Start by reading the linked FastBroadcast.jl section around lines 305-313 and the ArrayPartition broadcast style in RecursiveArrayTools.jl. Trace why the threaded fast_materialize path is skipped, then determine which repository owns the fix and verify threaded execution for a DynamicalODEProblem using the reported broadcast form.