fix(overlay): stop dropping members on multi-target actions

[char0n]

Updates the apidom dependency family to the latest releases.

package	before	after
@speclynx/apidom-core	^5.0.1	^5.1.1
@speclynx/apidom-ls	^2.11.6	^2.11.7
@speclynx/apidom-overlay	^5.0.1	^5.1.1
@speclynx/apidom-reference	^5.0.1	^5.1.1

Why

apidom 5.1.1 carries the fixes for speclynx/apidom#459 (style-preserving serializers drop shared element instances) and speclynx/apidom#460 (applyUpdateAction merges the same update element into every matched target). Together they resolve #127.

Verification

Reproductions were run against a sandbox pinned to the entire apidom family at 5.1.0 — the versions that would have resolved when #127 was filed — and compared against this branch.

Minimal case — $.components.schemas['ActorType', 'CardinalityFilterType'] with a single update:. On 5.1.0 the second match serializes as ? null explicit keys; on 5.1.1 both matches are correct. The JSON output path (which produced quiet null values rather than an obvious syntax oddity) and the copy: case where the source node sits later in the file are both clean as well.

Real FDX spec — FDX 6.4.0 shared components (OpenAPI 3.1.0, with genuine x-fdx-csdf-* extensions), with a reconstructed CSDF filter action:

	corrupted ? null members	intact x-fdx-csdf-technical: true
apidom 5.1.0	16	24
apidom 5.1.1	0	40

23 pre-existing in source + 17 matched = 40. On 5.1.0 only the first match survived and 16 were lost. Output validates cleanly (9 warnings, all pre-existing example deprecations).

On the open question in #127

The reporter's output showed both ActorType and CardinalityFilterType corrupted, while our reduction only ever corrupted the second-and-later match. That is now accounted for and needs no further fix.

The cause was a single WeakSet in toYAMLNode that was added to on entry and never unwound on exit, making it visit-once-ever rather than cycle detection. Eight variations were swept (union of 3, reversed union order, unmatched earlier schema, duplicate prior action, wildcard-then-union, nested update values) and the first match survived in every one — so no variation of the minimal example can corrupt both.

What does reproduce their exact output is an action that also matches a schema sitting earlier in the file than ActorType: the surviving copy lands on that earlier schema, leaving both pasted schemas as ? null. Their overlay's other CSDF actions would do exactly this, and the first match simply wasn't in the excerpt they pasted. This is confirmed on the real FDX spec above, where ActorType itself comes out as ? null pre-fix.

Their five real overlay files remain unobtainable — nothing was ever attached to the issue — but the mechanism is fully reproduced on real FDX data, so they are no longer needed to validate the fix.

Testing

tsc --noEmit clean; npm test 82 passing.

Closes #127

[char0n]

🎉 This PR is included in version 1.4.1 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀