feat(vba-extractor): model DoCmd.OpenReport and DoCmd.OpenQuery like OpenForm (closes #48)

[ardelperal]

Closes #48.

What

Hueco 6 (DoCmd.OpenForm "X" -> opens-form edge, hueco B4) was always labelled follow-up work for OpenReport and OpenQuery. This PR lands both. The OpenForm pipeline is preserved byte-identically (existing tests stay green without modification); OpenReport mirrors it through a small dispatch table; OpenQuery opts for an UnresolvedReference so the resolver binds to the REAL query node emitted by SqlQueryExtractor.

Why

In a real Dysflow-managed Access project, every DoCmd.OpenReport "InformeMensual" and DoCmd.OpenQuery "ConsultaDepuracion" line produced zero graph edges. Cross-report and cross-query UI traffic was invisible to the resolver, even though .report.txt and queries/*.sql are first-class files in the index. This PR closes that gap for OpenReport (synthetic report-layout stub + heuristic edge) and OpenQuery (UnresolvedReference resolving to the existing query node).

Diff

File	Lines
src/extraction/vba-extractor.ts	+264 / −69
__tests__/extraction-vba.test.ts	+236 / −0
src/types.ts	+12 / −0
Total	+512 / −69

Inside the 400-line source-budget; the PR is reviewable as ONE unit (one logical refactor + two new emissions sharing one pipeline).

Design

OpenForm + OpenReport: shared pipeline via dispatch table

New DOCMD_OPEN_DISPATCH constant (a ReadonlyArray<...> of typed entries) carries everything scanDoCmdOpenCalls + emitOpensStubEdge need:

{
  method, re,
  edgeKind, stubKind,
  syntheticPrefix, syntheticExtension,
  moduleNamePrefix, cacheKey,
  metadataTargetKey, synthesizedBy,
}

• OpenForm: edgeKind opens-form, stubKind form-layout, synthetic prefix synthetic:opensFormStub, extension .form.txt, qualifiedName prefix Form_, metadata key targetFormName, tag vba-opens-form.
• OpenReport: edgeKind opens-report, stubKind report-layout, synthetic prefix synthetic:opensReportStub, extension .report.txt, qualifiedName prefix Report_, metadata key targetReportName, tag vba-opens-report.
• OpenQuery is intentionally NOT in this dispatch — see below.

The scan/emit pipeline iterates the dispatch; OpenForm's emit is byte-identical to pre-#48 (same regex, same arg resolution, same stub id formula, same metadata key — pin tests stay green). The cache key is ${cacheKey}:${loweredName} so OpenForm and OpenReport de-dup buckets stay disjoint within a file. The synthetic path's file extension mirrors the real file extension per Issue #48 spec.

OpenQuery: UnresolvedReference pattern (NOT a stub + edge)

Following the vba-me-control and vba-forms-bang precedent. scanDoCmdOpenQuery is a separate scanner (not in the dispatch table) that pushes ONE UnresolvedReference per match:

• referenceName = resolved query name
• referenceKind = 'references'
• metadata.synthesizedBy = 'vba-opens-query'
• NO synthetic function or query node — SqlQueryExtractor already emits the real query node for queries/<Name>.sql, and the resolver binds the reference to it via name match.

Type system

NodeKind extended with 'report-layout'. EdgeKind extended with 'opens-report'. Distinct from form-layout / opens-form so downstream tooling can filter form vs report without inspecting qualifiedName.

Test coverage

9 new atoms in __tests__/extraction-vba.test.ts (Issue #48 describe block):

OpenReport (5 atoms):

1. Literal "InformeMensual" → emits opens-report edge + report-layout stub
2. Const-resolved REPORT_MENSUAL + unknown REPORT_UNKNOWN (mirrors OpenForm Const-fallback)
3. Stub id deterministic across re-index (asserted against generateNodeId('synthetic:opensReportStub/<Name>.report.txt', 'report-layout', '<Name>', 0))
4. W4 no-synthetic-function-node guard
5. De-dup: two calls produce exactly ONE stub

OpenQuery (4 atoms):
6. Literal "Consulta1" → ONE UnresolvedReference, no synthetic nodes
7. Const-resolved CONSULTA_DEPURACION → UnresolvedReference with resolved name
8. Const-unknown CONSULTA_UNKNOWN → falls back to bare identifier
9. W4 no-synthetic-function-node guard for OpenQuery too

Regression guards (existing tests, all green without modification):

• extraction-vba-control-modeling.test.ts hueco-6 test
• extraction-vba.test.ts "DoCmd.OpenForm does not emit a synthetic function node" (W4)
• extraction-vba.test.ts "resolves local string constants in DoCmd.OpenForm and falls back..." (Const-fallback)

Validation

• pnpm exec vitest run __tests__/extraction-vba.test.ts -t "Issue #48" → 9/9 pass in 681 ms
• Full VBA suite (6 files: extraction-vba, extraction-vba-control-modeling, extraction-vba-form, extraction-vba-enums-consts, extraction-vba-realfixtures, extraction-vba-roadmap-25-26) → 236/236 pass in 5.82 s — zero regressions in OpenForm or any other extractor path
• pnpm run build → exit 0, no TS errors

Out of scope (intentional)

• OpenTable and other DoCmd.Open* siblings. Same dispatch pattern is ready to absorb them later; not adding them now keeps the diff focused.
• Arbitrary variable data-flow for OpenForm/OpenReport/OpenQuery arguments — bare identifiers resolve only through local Const declarations, consistent with the existing OpenForm scope.

Not done

n/a — issue complete in this PR.