Fix corner index ordering

[willirath]

Description

Fix corner index ordering in the A-grid and C-grid interpolators: the flattened gather currently in main silently produced wrong corners when particless time or depth indices were not aligned, so a particle could pick up a batch-mate's ti/zi.
Both interpolators now go through a single and tested _gather_corners helper that derives index arrays and output shape from the same layout.

Adds a regression test that a trajectory does not depend on its batch-mate release times, plus interpolation tests for mixed per-particle indices.

Also updates the expected z values in test_nemo_3D_curvilinear_fieldset[RK4_3D], which had recorded the buggy mixing.

Reproducer with U uniform in space, varying in time, so each particle's displacement depends only on its own clock:

# fieldset: flat A-grid, U = cos(2*pi*t/86400) m/s (uniform in space), V = 0, 3-hourly
t0 = np.timedelta64(0, "s")

def run(release_times):
    npart = len(release_times)
    pset = ParticleSet(
        fieldset, pclass=Particle,
        t=np.array(release_times),
        z=np.zeros(npart), y=np.zeros(npart), x=np.zeros(npart),
    )
    pset.execute(AdvectionRK4, dt=np.timedelta64(1, "h"), endtime=np.timedelta64(48, "h"))
    return pset.x[0]

alone     = run([t0])
staggered = run([t0] + [t0 + np.timedelta64(3, "h")] * 3)

assert staggered == pytest.approx(alone)  # fails before the fix

Particle 0 should be identical in both runs as only its batch-mates differ. Before the fix, staggered moves particle 0 onto the other particle time index and the assert fails. The depth analogue is the same shape: two particles at different z in one set get each other's zi.

Checklist

• Tests added: Integration test and test for the new corner gather logic.
• This PR targets the correct branch: Only relevant for v4. So targeted at main.

AI Disclosure

• This PR contains AI-generated content.
  • I have tested any AI-generated content in my PR.
  • I take responsibility for any AI-generated content in my PR.
  • Describe how you used it (e.g., by pasting your prompt): Bug was found by Claude Opus when developing a downstream package. Diagnostics and implementation of the fix were done with Claude Opus. I guided the session and reviewed the code.

Draft PR because it's late and I want to have a fresh pair of eyes (my own or someone else's) tomorrow.

[erikvansebille]

Great that you spotted this bug @willirath, and thanks for the fix! I had a quick skim and I think your problem diagnosis and solution are sound. Let me know when you want me to do a full/thorough review