test: tighten fusion-era assertions + PathSummary fusion correctness suite

[JohannesLichtenberger]

Summary

During the iter#28→32 fusion rollout (PRs landing the OBJECT_NAMED_* family) several existing tests were relaxed to accept either the legacy OBJECT_KEY shape or the post-fusion OBJECT_NAMED_* shape. Now that fusion is default ON, the legacy branch is unreachable — those OR/role-predicate assertions are silently weaker than they look. Per the project rule "do not relax tests," tighten them back to the exact fused kind dictated by each test's JSON shape.

Also add a new PathSummaryFusionTest because the user asked specifically: make sure that with fused nodes, the PathSummary is still valid.

Tightenings (3 files)

JsonNodeTrxCopyTest.java

4 assertion sites switched from assertTrue(kind.playsObjectKeyRole(), …) (accepts any of the 6 fused kinds) to assertEquals(NodeKind.OBJECT_NAMED_<TYPE>, kind) per JSON value type. Removed the unreachable legacy if/else descent branches whose else-arm walks an extra OBJECT_STRING_VALUE/etc. child that fusion no longer emits.

FlyweightCursorTest.java

• 5 assertion sites tightened the same way (OBJECT_NAMED_STRING for string-valued fields, OBJECT_NAMED_NUMBER for numeric).
• Descendant-count assertion was assertTrue(count >= 3, …) (accepts both the 3-fused and 5-legacy answer); now assertEquals(3, count, …).

LdjsonShredderTest.java

2 sites at testTreeStructure tightened to OBJECT_NAMED_STRING and now also assert the inline value (\"v\" / \"v2\") — a legitimate strengthening since the fused leaf carries the value directly.

New: PathSummaryFusionTest (6 tests, all passing)

Pins down the contracts in PathSummaryWriter:

1. fusedPrimitiveLeaves_registerCanonicalOBJECT_NAMED_OBJECT_pathKind — {s,n,b,z} of all four primitive flavours each produces one level-1 entry under the canonical OBJECT_NAMED_OBJECT pathKind with references=1.
2. duplicateFieldName_acrossSiblingObjects_increments_references — [{name:\"a\"},{name:\"b\"}] collapses into one path entry with references=2.
3. removingFusedField_decrements_pathSummaryReferences — removing one of two fused OBJECT_NAMED_NUMBER records drops refs from 2 → 1.
4. removingLastFusedField_collapses_pathSummaryEntry — removing the sole instance of a field eliminates the entry entirely (no orphan refs=0 entry).
5. fusedStructural_OBJECT_NAMED_ARRAY_registers_pathSummaryEntries_onBothLevels — Phase 2 structural fusion of {items:[1,2,3]} mirrors at both the field-name level (OBJECT_NAMED_OBJECT pathKind for items) AND the underlying array level (ARRAY pathKind for elements).
6. renamingFusedField_migrates_pathSummaryReferences — setObjectKeyName(\"oldName\" → \"newName\") migrates the reference; old entry gone, new entry refs=1, registered under the canonical pathKind.

Test plan

• ./gradlew :sirix-core:test --tests \"io.sirix.access.node.json.JsonNodeTrxCopyTest\" — 14 tests, 0 failures.
• ./gradlew :sirix-core:test --tests \"io.sirix.access.trx.node.FlyweightCursorTest\" — 15 tests, 0 failures.
• ./gradlew :sirix-core:test --tests \"io.sirix.service.json.shredder.LdjsonShredderTest\" — 23 tests, 0 failures.
• ./gradlew :sirix-core:test --tests \"io.sirix.access.node.json.PathSummaryFusionTest\" — 6 tests, 0 failures.

Notes

• Five other tests (HOTSplitIntegrationTest, HOTVersioningIntegrationTest, JsonStructuralFuzzTest, HOTHeightOptimalTest, HOTIndexManyRevisionsTest, etc.) use playsObjectKeyRole() as dispatch (if (kind.playsObjectKeyRole()) { … }), not as an assertion. Those are legitimate flow-control predicates, not relaxations — left alone.
• ProductionReadinessTest uses playsObjectKeyRole() inside walker dispatch and one comment says "accept either kind via NodeKind#playsObjectKeyRole()." That comment is stale (fusion default ON), but the predicate is still a dispatch usage, not a weakened assertion. Left alone.