Print more details when regression tests fails

[Luthaf]

This should help debug failures when running the same test on different CI runners

Contributor (creator of pull-request) checklist

• Tests updated (for new features and bugfixes)?
• Documentation updated (for new features)?
• Issue referenced (for PRs that solve an issue)?

Reviewer checklist

• CHANGELOG updated with public API or any other important changes?

[Luthaf]

On some runners (https://github.com/metatensor/metatomic/actions/runs/36716334056/job/109891461253), this showed failures like

===== mismatches for 'mtt::cs_iso' (rtol=1e-04, atol=1e-04) =====
key=(o3_lambda=1, o3_sigma=-1) 1/24 elements not close, max abs diff=0.0004
    sample=(system=0, atom=7) components=[(o3_mu=-1)] property=(shielding=0): actual=2.85514 expected=2.85554 diff=0.0004

And

===== mismatches for 'energy' (rtol=1e-06, atol=1e-05) =====
key=(_=0) gradient 'strain' 2/9 elements not close, max abs diff=1.3e-05
    sample=(sample=0) components=[(xyz_1=0), (xyz_2=2)] property=(energy=0): actual=-1.79819 expected=-1.79821 diff=1.3e-05
    sample=(sample=0) components=[(xyz_1=2), (xyz_2=0)] property=(energy=0): actual=-1.78524 expected=-1.78525 diff=1.3e-05

In all cases, this seems to be coming from the fact that we run the models in float32 with fairly tight tolerances, so the fix is to loosen the tolerances a bit.

[EricBoittier]

LGTM

[EricBoittier]

Nice this will be very valuable going forward.