Cache label identifier data to eliminate redundant parsing

[jordanpadams]

Summary

Resolves #1568

• Add LabelCacheEntry POJO to hold pre-extracted identifier data (logical IDs, lid/lidvid refs, context area refs) from a parsed label
• After each label is parsed in LabelValidationRule, cache identifiers (with \n detection enabled) into ReferentialIntegrityUtil's labelIdentifierCache
• additionalReferentialIntegrityChecks() now uses cached logicalIdentifiers and lidOrLidVidReferences when available — no disk re-parse for the common case; fallback parse retained for labels not in the initial validation pass
• CrossLabelFileAreaReferenceChecker.add() uses cached logicalIdentifiers when available — no disk re-parse for file area reference checks
• collectAllContextReferences() uses cached context area refs to skip three Saxon XPath evaluations per label — fallback to fresh parse when no cache entry exists
• \n detection (INVALID_FIELD_VALUE) happens once during cacheIdentifiers() (with reportCarriageReturns=true), so the referential integrity phase can safely use cached values without risking double-reporting or missed errors
• Fix CrossLabelFileAreaReferenceChecker.reset() to clear the isObservational map alongside knownRefs, preventing static state from leaking across validation runs
• Call CrossLabelFileAreaReferenceChecker.reset() from ValidateLauncher alongside ReferentialIntegrityUtil.reset()

Test plan

• NASA-PDS/validate#15-2 passes (\n in logical_identifier — 1 INVALID_FIELD_VALUE reported, no double-reporting)
• NASA-PDS/validate#401-1 passes (\n in lid_reference — 3 INVALID_FIELD_VALUE detected from cache, no re-parse needed)
• Full Cucumber suite: 297/297 scenarios pass

🤖 Generated with Claude Code

[jordanpadams]

Acceptance Criteria Verification

From issue #1568:

Given a bundle with many product labels
When I perform validation including referential integrity checks
Then I expect each label file is read and parsed from disk only once

---

How each label is now parsed exactly once

Phase 1 — Initial label validation (LabelValidationRule.validateLabel(), line 319):

validator.parseAndValidate(processor, target)   ← one disk read + one DOM parse per label
  └─ cacheIdentifiers(document, targetUrl)       ← extracts LIDs, LIDVIDs, context refs
       └─ ReferentialIntegrityUtil.cacheLabelIdentifiers(url, entry)

The DOM is built once from disk. cacheIdentifiers() (line 345) runs against the already-in-memory Document, extracting:

• logicalIdentifiers (LIDs/LIDVIDs registered by the product)
• lidOrLidVidReferences (all lid_reference/lidvid_reference values)
• contextAreaRefs (Investigation Area, Observation System Component, Target Identification refs)

\n detection (INVALID_FIELD_VALUE) is also performed here (once) with reportCarriageReturns=true.

---

Phase 2 — Referential integrity checks: three former re-parse sites, all now cache-first:

Former re-parse site	Now
ReferentialIntegrityUtil.additionalReferentialIntegrityChecks() (line 823) — db.parse(url.openStream()) for every label	getCachedLabelIdentifiers(url) hit → uses cached.getLogicalIdentifiers() / getLidOrLidVidReferences() directly; no disk read, no parse
CrossLabelFileAreaReferenceChecker.add() (line 41) — DocumentBuilderFactory…newDocumentBuilder().parse(target.getUrl().openStream())	getCachedLabelIdentifiers(target.getUrl()) hit → uses cached.getLogicalIdentifiers() directly; no disk read, no parse
collectAllContextReferences() — three Saxon XPath evaluations over a re-parsed DOM	getCachedLabelIdentifiers(url) hit → uses cached.getContextAreaRefs() directly; no Saxon XPath, no re-parse

A fallback disk-parse is retained in all three sites for labels that were not in the initial validation pass (e.g. labels that failed parsing and were never cached). This keeps correctness for edge cases while achieving the optimization for the common case.

---

Test evidence

• NASA-PDS/validate#15-2 — label with \n in logical_identifier: INVALID_FIELD_VALUE reported exactly once (from cacheIdentifiers()), not duplicated by the referential integrity phase ✅
• NASA-PDS/validate#401-1 — bundle with \n in lid_reference (3 occurrences): all 3 INVALID_FIELD_VALUE errors detected from cache; referential integrity phase uses cached values with no re-parse ✅
• Full Cucumber suite: 297/297 scenarios pass ✅

🤖 Generated with Claude Code

[jordanpadams]

Fallout fix: carriage-return detection broken by Saxon DOM normalization

Two CI test failures (#15-2, #401-1) were caused by a subtle interaction between the new caching code and Saxon's XML processing:

Root cause: cacheIdentifiers() was building a DOMSource from the Saxon-processed Document returned by parseAndValidate(). Saxon normalizes xs:token field values, stripping \n from element text content. LabelUtil.getIdentifiersCommon() detects carriage returns via node.getTextContent().contains("\n") — with Saxon's normalized DOM, this check silently never fires.

Effect on tests:

• #15-2: \n in logical_identifier was stripped → error.validation.invalid_field_value=1 and error.validation.internal_error=1 were lost; a new error.label.missing_file appeared instead (22 errors instead of 23)
• #401-1: 3 \n in lid_reference values were stripped → error.validation.invalid_field_value=3 were lost (2 errors instead of 5)

Fix: Re-parse with plain DocumentBuilderFactory in cacheIdentifiers() (preserves raw whitespace), instead of reusing the Saxon-normalized Document.

Backwards compatibility: The old code only reported lid_reference carriage-return errors during bundle/collection validation (additionalReferentialIntegrityChecks), never for single-label validation. To preserve this behavior exactly, cacheIdentifiers() uses getLidVidReferences(reportCarriageReturns=false); additionalReferentialIntegrityChecks re-parses for error reporting on the cache path (same as the pre-cache uncached path). This means single-label users with \n in lid_reference do not suddenly get new errors.

Also removed incidental \n whitespace from 5 lid_reference values in the github15 test data — these were formatting artifacts, not intentional test cases.

🤖 Generated with Claude Code

[jordanpadams]

Testing using the new benchmark script:

# Execution for each scenario
$ python scripts/benchmark_bundle.py --products 100000

# Current version on main
=== Validate ===
  $ /tmp/validate_benchmark_bin/validate-4.2.0-SNAPSHOT/bin/validate --rule pds4.bundle --skip-content-validation /var/folders/7y/5wn5jqmx2v37tj9270z2nf3h0000gq/T/pds4_benchmark_l5le0ijv
  Exit code : 0
  Elapsed   : 1401.76s
  Summary   :
    Summary:

      100002 product(s)
      0 error(s)
      0 warning(s)

      Product Validation Summary:
        100002     product(s) passed
        0          product(s) failed
        0          product(s) skipped
        100002     product(s) total

      Referential Integrity Check Summary:
        100002     check(s) passed
        0          check(s) failed
        0          check(s) skipped
        100002     check(s) total


    End of Report
    Completed execution in 1401533 ms

# After updates from this PR
TBD