Fix clear_module! world semantics

[vtjnash]

The isconst is a world-specific query, so it needs invokelatest too. Restructure a bit of the code to do that in a helper function. Also add a call to delete all bindings in v1.12+. Although that may be expensive and currently warns that it may be incorrect, it may be helpful to get testing of that.

[fingolfin]

Looks good to me, thank you!

[fingolfin]

If there is a chance for this to be slow, I'd definitely would want to check this in the Oscar manual were a slowdown in this function caused us major pain before.

[fingolfin]

Hmm, fails Julia 1.6 tests (finalizer of an object in the cleared module is not called).

(and of course a changelog entry is missing)

[fingolfin]

Building Oscar.jl docs with Documenter.jl master ...

• ... and Julia 1.10 takes 35 secs
• ... and Julia 1.11 takes 45 secs
• ... and Julia 1.12 takes 47-53 secs

So it seems Julia 1.12 is getting slightly slower than 1.11, which is a bunch slower than 1.11 (unfortunately that's a general pattern I am seeing and a reason why I do my daily work with Julia 1.10 still).

On the bright side, Base.delete_binding doesn't seem to matter much here (I tried commenting that line out and measured no difference).

[fingolfin]

@vtjnash would you mind adding a changelog entry?

[mortenpi]

Thank you!

If this is a blocking bug somewhere (Base manual?) then we could get this out in a backport to 1.10 quickly. Otherwise, there are a few more PRs we might want to get into Documenter 1.11.0.

[vtjnash]

Anytime in a week or two is fine and will hold the bugfix upstream. Not urgent