Non-deterministic CI results for `VolumeIntegralAdaptive` with `VolumeIntegralSubcellLimiting`

Somehow, we have failing test for `elixir_euler_sedov_adaptive_sc_subcell.jl` from time to time.
See for instance the [runs of this PR](https://github.com/trixi-framework/Trixi.jl/actions/runs/26823926280/job/79145797799) where the first run failed but the second passed.
The errors are about `1e-3` to `1e-4`.

We already noticed random failing test in #2924 where `VolumeIntegralAdaptive` was extended to `VolumeIntegralSubcellLimiting`. #3009 couldn't fix it. Next try in #3056.

Hot fix (increasing test tolerance) in #3061 .

Possible reason: Subcell limiting is known to be very sensitive to small changes. So, small differences can accumulate to errors of `1e-3`. Additionally, `VolumeIntegralAdaptive` has a hard switch to decide whether to use the default of the stabilized volume integral. That is known. Are there other reasons? Bugs?
Do we have the same problem with different julia versions than 1.10?

Start by repeatedly running elixir_euler_sedov_adaptive_sc_subcell.jl and recording the failing values. Then inspect VolumeIntegralAdaptive and VolumeIntegralSubcellLimiting, review the attempts in #2924, #3009, and #3056, and compare behavior across Julia versions. Done means the source of the nondeterminism is identified and a reliable regression test or fix is established.

Written by the indexing model from the issue text.