Revised and cleaned testing environment

[Da-Be-Ru]

Splitting the package and testing environments with separate Project.tomls created inconsistencies and compat problems that caused the testing environment to lag behind upstream changes to dependencies (specifically LinearSolve in this case).

As suggested by @pjaap, I pushed the testing environment dependencies into an [extras] section to keep it better organized and covered in one sweep by the CompatHelper and Aqua etc. (see related issue in ExtendableFEM).

[Da-Be-Ru]

Some tests are expected to fail as they now reveal incompatibilities (e.g. in LinearSolve) for packages that previously remained frozen by the compat restrictions.

Finally, this change also addresses the downstream issue of the VoronoiFVM test environment not instantiating correctly (see related issue).

[j-fu]

The failing tests are due to a problem in LinearSolve which is currently being fixed, see SciML/LinearSolve.jl#591 . I think we wait for this to be merged.

[j-fu]

For the test environment stuff, see https://discourse.julialang.org/t/special-environment-for-testing/90306 and https://pkgdocs.julialang.org/v1/creating-packages/#adding-tests-to-packages

While test/Project.toml is the newer way how to handle this, it seems that currently, the choice between [extras] and test/Project.toml appears to be seen as a matter of taste. I personally prefer the later (as you have seen...) although it also has a bit of (black) magic. In particular it was easier to add some package to the test environment via Pkg after activating it. But with 1.11, we have Pkg.add("XPkg", target=:extras), so that argument evaporated...

[ChrisRackauckas]

We were told by Pkg folks that the latter has issues and we should prefer the [extras] form.

[pjaap]

CI is fine now.

[pjaap]

Is anything missing? If not, let's merge and register a new version. @Da-Be-Ru please add yourself to the authors and increase the patch version number to 1.7.1.

[j-fu]

But see also Da-Be-Ru#1 , could be done with this PR as well.
Not sure if we shouldn't also bump the compat of LinearSolve to "2.36.0, 3.7.1" .
EDIT: we should, and indeed this would be the only formal reason to register 1.7.1, as reorganization of tests doesn't affect the API.

[Da-Be-Ru]

I'd do the honours, but I made some last changes, so I'd rather someone approve of the final state before merging 👀

[j-fu]

LGTM (looks good to me)