Add trim tests

[RomeoV]

Add a trim directory with basic tests for trimming.
I tried to imitate the code/package loading structure of SciMLBase.

There are a few quirks here, most of which hopefully will resolve themselves soon:

1. We need to dev seven packages to appease JET and juliac. Each package's reason is documented in the new test/trim/Project.toml file. Also, when the upstream branches are merged and deleted, this will fail. But that's good so we can bump the compat accordingly.
2. The testing code includes clean_optimization.jl and trimmable_optimization.jl. One day hopefully both will be trimmable. At the moment both versions pass JET (impressive!) but only the heavily preallocating version passes juliac.
3. In order to run juliac we have to put the code to trim into a package. It seems a module will not do. This is documented in the src/TrimTest.jl file.
4. This is all Julia 1.12.0-rc1 only. But the runtests.jl should take care of that.

[RomeoV]

Ready for review.

[RomeoV]

I think the two main tests to look here are:

• pre, ubuntu-latest, GROUP=trim
• pre, macos-latest, GROUP=trim

I don't think anything else should have been impacted...

[RomeoV]

The tests are looking good. Checking the CI results from the previous results, everything that passes on master also passes here. Most importantly, Ubuntu w/ julia-pre, which is the only one that should actually be affected (besides MacOS julia-pre, which crashes for an unrelated reason).
I just pushed one more commit fixing the formatting, so the whole CI has to run again, but otherwise I'd say it's ready to merge.

[ChrisRackauckas]

We shouldn't merge this with stuff still on branches. Need to get the upstreams fixed first.

[topolarity]

This looks like it's worth a bug report upstream

[ChrisRackauckas]

Claude wasn't able to reproduce it with an MWE, so @RomeoV can you take this?

[ChrisRackauckas]

This should get updated to CPUSummary v0.2.7 which should fix the Polyester and PolyesterWeave pieces (i.e. remove those bits, and trim should work still just with release)

[ChrisRackauckas]

@aviatesk can you look into this?

[ChrisRackauckas]

@AayushSabharwal was this completed?

[AayushSabharwal]

No, this is running into the JET failures in LinearAlgebra/@assert in ForwardDiff I mentioned in slack. It fixes the type instability relevant here, we just can't test it outside of 1.11.

[RomeoV]

I'm trying to wrap the const cache = init(...) calls into a OncePerProcess as suggested by Jeff in JuliaLang/julia#59215 (comment). However, I am getting this error during trimming

MethodError(f=Base.Enums.namemap, args=(SciMLBase.ReturnCode.T,), world=0x0000000000009661)

@ChrisRackauckas I recall that we got this error during the hackathon at JuliaCon. I recall we tried to put something like

println(Core.stdout, SciMLBase.ReturnCode.T)

into the precompile, but it doesn't seem to help, unfortunately. Do you have any other ideas?