feat(query): implement scalar functions and XML Schema casts (#563)

[remiceres]

Summary

Implements #563 in the native core.next.query pipeline: SPARQL scalar built-ins, XML Schema constructors and SPARQL expression type-error handling.

• Enable scalar SELECT expressions through the existing AST/compiler and KGRAM projection machinery; aggregate projections remain explicitly unsupported.
• Implement the seven standard XML Schema casts and ISNUMERIC, validate numeric lexical forms/value spaces, preserve source RDF terms and handle exact integer truncation, signed zero and canonical numeric-to-string casts.
• Complete string/language, Unicode substring, URI, BNODE scoping, date/time, numeric and hash behavior. Accept empty CONCAT/COALESCE argument lists in the existing ANTLR grammar; no parser dependency is added.
• Distinguish absorbable SPARQL type errors from engine/infrastructure failures in FILTER, BIND, projection, IF, COALESCE and boolean expressions.
• Add 118 regression cases, including source-term preservation for all four disputed upstream cast-decimal bindings.

Validation

• ./gradlew test: 2,934 tests, zero failures/errors/skips.
• SonarLint: 20/20 changed Java files analyzed, zero diagnostics.
• Full sibling W3C run: all 73 harness unit tests pass; all active non-SPARQL suites pass. EARL parsing, 15 validation queries and JSON/EARL coverage checks pass.
• Same 978 official SPARQL test identifiers compared against the pre-[Query] Implement SPARQL 1.1 scalar built-in functions and casting operators #563 baseline: 644 → 738 passed (+94), zero regressions, no missing tests with the companion harness fix. Core-only with the unchanged harness gives 730 passes.
• SPARQL 1.0: 398 → 415/483. SPARQL 1.1: 246 → 323/495.
• functions: 75/75 pass. Cast evaluation: 5 pass + 1 indeterminate; all seven cast syntax tests also pass.

The 239 remaining SPARQL failures are pre-existing cases outside this issue. The additional indeterminate case is not claimed as passing: upstream cast-decimal.srx rewrites four directly projected source literals. The companion harness PR continues executing it, recognizes only that exact discrepancy and preserves the original JUnit failure. No official fixture is changed and no core behavior is tailored to its erroneous expected source terms.

Integration order

Merge this PR into feature/corese-next first, then merge corese-stack/corese-w3c#27. The harness uses the existing native parser only to identify computed numeric projections; numeric value comparison remains independent of Corese's numeric implementation. Its scoped comparison is justified by the upstream discussion, without relaxing directly projected RDF-term identity.

[github-actions]

Overall Project	54.2% -0.08%	🍏
Files changed	82.51%	🍏

File	Coverage
QueryTypeErrorException.java	100%	🍏
NativeExpressionEvaluator.java	95.7%	🍏
CoreseAstQueryBuilder.java	93.31%	🍏
ASTConstants.java	92.45%	🍏
NativeStringExpressionEvaluator.java	91.88% -3.35%	🍏
SparqlExpressionBuilder.java	86.27%	🍏
NativeCastExpressionEvaluator.java	81.87% -18.13%	🍏
NativeEvaluationContext.java	81.66% -7.96%	🍏
NativeNumericValues.java	72.06% -27.94%	🍏
NativeNumericExpressionEvaluator.java	70.94% -2.66%	🍏
SparqlTermResolver.java	68.59%	🍏
IsNumericAst.java	66.67% -33.33%	🍏
NativeTemporalExpressionEvaluator.java	62.89% -5.03%	🍏
NativeBooleanExpressionEvaluator.java	62.28% -0.6%	🍏
NativeIriExpressionEvaluator.java	61.76% -6.86%	🍏
NativeValueComparison.java	44.94% -6.33%	❌
Eval.java	44.52%	🍏
Memory.java	43.57% -0.34%	🍏

[github-actions]

Test Results

  427 files    427 suites   25s ⏱️
2 915 tests 2 915 ✅ 0 💤 0 ❌
2 934 runs  2 934 ✅ 0 💤 0 ❌

Results for commit 4dd4a38.