feat: YAML reader for BattleScribe-schema data + file-based spec adapter setup

[amis92]

What

Foundation work for the muster CI harness (golden-roster regression checks for data repos):

• New WarHub.ArmouryModel.Source.Yaml: reads YAML-serialized BattleScribe data (the BSData/wh40k-11e format) into the node model via a YAML→XML structural transform feeding the existing serializer. Scalars flow as raw strings (representation model — no typed-deserialization coercion; pinned by test), strict mode throws on unknown collection keys, $text → element text, text-element keys (comment/description/customNotes/readme) emit child elements. Loads all 47 real wh40k-11e files (integration suite, gated by WH40K11E_DATA env var, skips cleanly when unset).
• SetupFromFiles on SpecRosterEngineAdapter (XML + YAML by extension), enabling dataSource-backed conformance specs and muster fixtures; parse failures name the offending file.
• 4 additive enum members found in real 11e data: ModifierKind.Replace/Ceil/Floor, ConditionGroupKind.Count (with XmlEnum attrs → full serializer round-trip). ⚠️ These are parse/serialize-only: no conformance specs exercise their evaluation semantics yet — follow-up needed before engine behavior on them is trusted.
• Release-build hygiene: fixed 16 pre-existing CA1822 (+ cascade CA1822/CA1859/CA1305 in StateMapper) that broke -c Release builds (CI builds Debug only, so these were invisible). Public API signatures unchanged (suppressed on public members, static-ified privates only).
• Spec submodule bumped to feat(testkit): HarnessError crash marker + YAML dataSource enumeration battlescribe-spec#307 (HarnessError + YAML dataSource enumeration).

Known follow-ups (not in this PR)

• associations/alias YAML keys load but round-trip lossily (no model support; documented).
• Roster-level cost aggregation drops cost types only reachable via entryLinks — real defect found by muster's pilot fixtures; filed separately with code-level analysis.

Verification

Source.Yaml.Tests 54/54 (with data) / green-with-skips (without); RosterEngine.Spec.Tests 31/31; ConformanceTests 351/351 (baseline preserved); Release builds of all touched projects clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KcYK8hmTmXi8eLWpg9LGKJ

[amis92]

Landed as #325, rebased onto the roster-engine stack (#315–#322) rather than onto feature/roster-engine, which is now closed.

Two changes during the rebase:

• The four enum additions were dropped as redundant. ModifierKind.Replace/Ceil/Floor and ConditionGroupKind.Count all reached main independently via feat: model NewRecruit schema additions over BattleScribe v2.03 #308, which modelled the NewRecruit additions in bulk. I verified they are a strict subset before dropping — main carries 22 ModifierKind members and 18 ConditionGroupKind members, a superset of everything this branch added, so nothing is lost.

  Your doc comments for Prepend and Replace were worth keeping, though — feat: model NewRecruit schema additions over BattleScribe v2.03 #308 covered the whole block with a single // comment. They are restored in docs: document string ModifierKinds and cover them with tests #326.

• ab77d59 ("resolve pre-existing CA1822 Release-build errors") was dropped automatically — git reported patch contents already upstream. I had independently hit the same 16 errors while splitting the stack and reached the same conclusion you did (public engine API stays instance-based; suppress per-member rather than make it static), so feat: add ISymbol-based roster engine core #321 already carries a byte-identical fix. Your diagnosis in that commit message was also what led me to check Release across every layer.

Everything else — the YAML reader, SetupFromFiles, and the submodule bump — came across unchanged and is on main.