Integrate pure Float32 computation checks into CI

[huiyuxie]

I have just refactored my package to support Float32 computation see trixi-gpu/TrixiCUDA.jl#94, so I need to ensure that all the Float32 computations are type stable here at least within rhs! functions (except for some unavoidable Float64 computations).

Type stability checks for single functions is already integrated in CI https://github.com/trixi-framework/Trixi.jl/blob/main/test/test_type.jl, but it is not enough to ensure the type stability for the whole computation process. I will try to integrate a new pure Float32 computation checks into CI like here #2129 (comment) by initializing the tests in an independent environment - but it may fail.

Can you @ranocha help review and give me some suggestions in the process - you clearly know all the problems and contexts.

[huiyuxie]

But this definitely add more workload to CI.

[ranocha]

Nice idea! Adding a few additional checks for this is fine - but we cannot test every elixir of course. Please feel free to ask me if you have some questions.