feat(vba-extractor): support bang operator - Me!Ctl, Forms!Form!Ctl (closes #44)

[ardelperal]

Closes #44.

What

Enables two Access VBA bang-operator idioms that the previous extractor silently dropped:

• Me!<Control> — the default-collection shortcut for Me.<Control>. Same emission shape (byte-identical referenceName, referenceKind, metadata.synthesizedBy), so the resolver needs no special-casing.
• Forms!<Form>[!<Ctl>] and Forms("<Form>")!<Ctl> — cross-form UI traffic. Previously doubly-invisible: Forms is in RUNTIME_RECEIVER_BLACKLIST so CALL_RE skipped it, and there was no dedicated scanner.

Why

In a real Dysflow-managed Access project (codegraph-vba issue #44), every cross-form Forms!FormPrincipal!txtEstado.Value = … line produced zero graph edges — so Form_FormPrincipal.cls and the Forms/* reference trail were unrelated after indexing. Same for Me!txtNombre inside form .cls bodies. The resolver layer alone couldn't recover these because the references didn't exist in the first place.

Diff

File	Lines
src/extraction/vba-extractor.ts	+162 / −13
__tests__/extraction-vba.test.ts	+163 / −0
Total	+312 / −13

Inside the 400-line review budget → single PR, no chain required.

Design

• Extend ME_CONTROL_RE from /\bMe\.(…)/gu to /\bMe[.!](…)/gu. The +3 column offset in scanMeControlReferences remains valid — both Me. and Me! are 3-character prefixes.
• New FORMS_BANG_RE with two alternatives: bang form (Forms!<Form>(?![.\w]), optional !<Ctl>) and paren form (Forms("<Form>")!<Ctl>, required trailing control). Negative lookahead on the bang form is what excludes Forms!FormX.Recordsource (property access, not control access).
• New scanFormsBang() method beside scanMeControlReferences; one UnresolvedReference per match (metadata.synthesizedBy = 'vba-forms-bang'). No synthetic function node — the form is a real .cls/.form.txt pair the resolver already picks up via vba-form-binding (W4 invariant).
• Wiring next to scanOpenFormCalls() — both need the original unmasked line because the paren form Forms("X")!Y has the form name inside a string literal.

Test coverage

8 new atoms in __tests__/extraction-vba.test.ts:

1. Me!txtFoo emits vba-me-control reference to txtFoo
2. Me!txtFoo and Me.txtFoo produce byte-identical UnresolvedReferences (parity regression)
3. Forms!FormX!txtY.Value = 1 emits vba-forms-bang reference to FormX
4. Forms!FormX (no control segment) emits vba-forms-bang to FormX
5. Forms("FormX")!txtY emits the same vba-forms-bang to FormX as the bang form
6. Forms!FormX.Foo (post-form property access) does NOT emit
7. rs!Campo (recordset field, STRETCH SCOPE) emits zero — pinned for future review
8. Forms!FormX!txtY does NOT synthesize a function node (W4 invariant)

Verification

• pnpm exec vitest run __tests__/extraction-vba.test.ts -t "Issue #44" → 8 passed in 569 ms
• pnpm exec vitest run __tests__/extraction-vba.test.ts __tests__/extraction-vba-control-modeling.test.ts __tests__/extraction-vba-form.test.ts __tests__/extraction-vba-enums-consts.test.ts __tests__/extraction-vba-realfixtures.test.ts __tests__/extraction-vba-roadmap-25-26.test.ts → 227 passed in 5.09 s (zero regressions across the entire VBA test surface)
• pnpm run build → exit 0 (tsc + asset copy OK)

Out of scope (intentional)

• rs!Campo (DAO/ADO recordset default-member field access). The runtime-receiver blacklist plus the absence of any Forms/Me prefix means the existing scanners already skip it. Pinned by a test so a future change can be reviewed explicitly against the bang-form scope decision.
• <Form>::<Ctl> qualified control emission — the bang scanner emits only the form reference today; the control segment is consumed but not emitted. Control emission is the form's responsibility downstream.

Not done

n/a — issue complete in this PR.