Update adapt vs ref24 cost script for current APIs

[ss2098]

This PR updates scripts/adapt_vs_ref24_cost.py so the cost comparison script runs with the current Underworld3 APIs.

Changes include:

• convert AdvDiffusionSLCN estimate_dt() output to a scalar before passing it to solve()
• replace internal mesh._deform_mesh() calls with the public mesh.deform() API
• replace the retired smooth_mesh_interior(..., method="anisotropic") path with uw.meshing.node_redistribution()
• add fail-fast SNES convergence checks so the script does not continue after failed solves
• simplify the Stokes boundary condition so the cost diagnostic runs stably

This does not change solver internals.

Validation:

• pixi run python scripts/adapt_vs_ref24_cost.py

Example local output:

• res-24 plain step: 2.489 s
• res-16 plain step: 1.291 s
• a16s adaptation overhead: 2.562 s
• per-sim-time slowdown a16s / ref24: approximately 0.78x

[lmoresi]

Adversarial review

Same three findings as #476, and they bite harder here:

1. MAJOR — free-slip → no-slip on the upper boundary (in build()). This script's sole purpose is a cost comparison against a stored ref-24 baseline taken under free-slip with warm starts; the comparison is invalidated, not updated. Use add_rotated_freeslip_bc(0, "Upper").
2. MAJOR — zero_init_guess=True in the timed loops: cold-start timings against a warm-start baseline.
3. MINOR — deform() double-counts field transfer in the remap timings (it runs remesh_with_field_transfer internally).

Requested changes as on #476. The API updates themselves verified correct.