test(config): add grid-style silent-drop guards

[mrcsin]

Problem

The grid-style editor (GridStyleEditorViewModel) maintains ~80 style fields in three parallel blocks — property declarations, Seed, and BuildRecord — sitting 200+ lines apart in a 523-line file. A field surfaced in Seed but missing from BuildRecord's _source with { … } compiles fine and silently reverts on save, with no compile or test guard. Issue #118 will rewrite exactly these blocks (nest the flat record, type colors, split the view model), which is the highest-risk moment for re-introducing a silent drop. This lands the anti-regression net first.

What changed (tests only — no production change)

Three directional guards, split by direction and field kind so each catches what the others structurally cannot, all driven by one distinct-value fixture (GridStyleOptionsTestData.Distinct() — 77 unique, valid, in-range fields):

Guard	Catches
Seed-populates	a dropped/cross-wired Seed line — including the nullable (decimal?/FontFamily) and italic fields a naive round-trip is blind to
Perturbation (exact-one-field)	a dropped/mis-targeted BuildRecord line — a plain BuildRecord(Seed(x)) == x round-trip cannot (with keeps _source, which is the seed source)
Save→load (via facade)	a GridStyleDtoMapper drop / GridStyleMapper mis-wire — the flatten/unflatten layer the nest slice rewrites

A count == 77 assert guards the reflection filter against drift.

Why three guards, not one: the existing round-trip test seeds Default (32 identical #000000 colors, masks cross-wires) and structurally cannot fail on a dropped BuildRecord line, nor on a dropped Seed line for the 14 nullable-fallback fields (null → BuildRecord restores _source). The three directional guards close all of that.

Verification

• dotnet build SemiStep.slnx — 0 warnings; dotnet test — 1687 passed, 0 failed; dotnet format clean.
• The guards are proven non-vacuous by scratch checks (delete a line, confirm the matching test goes red, revert):
  • drop a BuildRecord line (InfoColor) → perturbation test red, names the field;
  • drop a nullable Seed line (ValidationPanelMaxHeight) → Seed-populates red (0 vs 150.5);
  • drop an italic Seed line → red (false vs true);
  • drop a GridStyleDtoMapper line (CellWeight) → save→load red (loaded 400 vs fixture 800).
• Review chain: comprehensive (5 agents) → smells → comment audit → critical, all clean after fixes. The reviews caught a real hole — the fixture's CellFontWeight: 400 equalled Default, and since that DTO field maps ?? default, a dropped mapper line would substitute 400 and leave the save→load guard green; changed to 800, and two reviewers independently re-verified all 78 fields now differ from Default.
• One residual, documented: a bool↔bool Seed cross-wire among the five all-true italics is inherently undetectable with a two-valued type — lower-probability than a dropped line.

This net unblocks the #118 debloat (slices 2–5: async Save, nest the record, type colors as StyleColor, split the view model into per-group drafts). #118 stays open.

Per-task commits before collapse

3638d1f docs(plan): record exec branch and verify steps for grid-style guards
96050d2 docs(test): trim restating comments in grid-style guards
547d222 test(config): name the surfaced-property count and fix Perturb weight derivation
2c8eeb4 test(config): fix fixture default-collision and trim redundant guards
373c6e0 test(config): add grid-style BuildRecord perturbation guard
471c4d2 test(config): add grid-style distinct fixture and value round-trip guards

🤖 Generated with Claude Code