feat(palace): expose adaptive mesh refinement controls

[Alisama20]

Summary

The generated config always carried a fixed Model.Refinement block with MaxIts = 0, so Palace's adaptive mesh refinement (AMR) could not be switched on from gsim. This adds RefinementConfig, which maps onto that block, and set_refinement() on every Palace simulation:

sim.set_refinement(max_its=5, tol=1e-3, max_dofs=2_000_000)

Palace's MaxSize is exposed as max_dofs, since it is a limit on the degrees of freedom and add_wave_port(max_size=...) already means something else.

Nothing changes by default. The block keeps the same three keys and values (UniformLevels: 0, Tol: 0.01, MaxIts: 0), which are also Palace's defaults, and keys left unset are not written, so Palace applies its own. uniform_levels is exposed too: it lets a coarse mesh file stand in for a finer mesh that Palace refines uniformly after loading.

In a frequency sweep, Palace refines on the error indicator averaged over the sampled frequencies (palace/fem/errorindicator.cpp), so a narrow resonance can end up under-refined. The docstring says so.

Part of #265.

Test Plan

New tests/palace/test_refinement_controls.py, following the pattern of #253. The key names and bounds are checked against Palace's own schema (scripts/schema/config-schema.json), which rejects unknown keys:

• the default block is unchanged, with and without a default RefinementConfig;
• every control reaches the config under Palace's key name, and only keys the schema knows are written;
• unset controls are left out;
• values outside the schema's bounds are rejected (Tol > 0, 0 < UpdateFraction < 1, non-negative integers);
• set_refinement() replaces the sim's config;
• write_config() forwards it, and the sim's write_config() passes it on. Removing either forwarding line makes exactly the matching test fail.

Also generated the full config.json of the simulation in nbs/palace_cpw_lumped.ipynb and validated it against the same schema with jsonschema: no errors by default, and none with every control set. The default file is byte-identical to the one main writes. As a control, the validator does reject a misspelled key, a negative MaxIts, UpdateFraction = 1 and Tol = 0.

Full suite: 1376 passed, 6 skipped (the 1363 that pass on main plus the 13 new tests). pre-commit: all hooks pass.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 64.73%. Comparing base (5b52244) to head (511242c).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #285      +/-   ##
==========================================
+ Coverage   64.67%   64.73%   +0.06%     
==========================================
  Files         107      107              
  Lines       16528    16554      +26     
  Branches     3271     3272       +1     
==========================================
+ Hits        10689    10716      +27     
+ Misses       4812     4810       -2     
- Partials     1027     1028       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.