Fix mesh filename from init steps in convergence tasks

[xylar]

Many convergence tasks have init steps with culled_mesh.nc, not mesh.nc as the final mesh file name. Here, we add a parameter to ConvergenceForward that specifies the mesh file name in the init step so this can be specified as appropriate.

The need for mesh.nc was introduced in #437 but it seems I rushed the review a little too much and didn't do enough testing to notice that the symlink was broken for many tasks.

This PR also fixes two small issues discovered when running the convergence suite:

• A trailing / in the name of the merry-go-round task paths is dropped for consistency with other task paths
• The convergence suite is updated to not require cached files that were seemingly never created for external_gravity_wave

Checklist

• Testing comment in the PR documents testing used to verify the changes

[xylar]

Testing

I ran the omega_pr (Omega) and convergence (MPAS-Ocean) suites on Chrysalis (with Intel).

Before this fix, the omega_pr suite was failing on various convergence tests. convergence hasn't been run in quite some time but surely would have shown failures as well.

/lcrc/group/e3sm/ac.xylar/polaris_0.9/chrysalis/test_20260102/convergence-fix-mesh-links
/lcrc/group/e3sm/ac.xylar/polaris_0.9/chrysalis/test_20260102/omega-pr-fix-convergence-mesh-links

omega_pr ran successfully. I saw state validation failures in the following tests in convergence:

0:00:02 FAIL ocean/spherical/icos/geostrophic/convergence_both
0:00:08 FAIL ocean/spherical/qu/geostrophic/convergence_both

This may indicate that the geostrophic test is not configured correctly (e.g. the time step is too large) but it does not seem to be related to this PR and reported in #443

[xylar]

@cbegeman, when you're back and have some time, please take a look.

[cbegeman]

Approving based on visual inspection and testing with Omega on chrys with intel, openmpi. Thanks, @xylar

[xylar]

Thank you for having a look, @cbegeman!