[JXPATH-204] Eagerly evaluate UnionContext before returning its value

[akashchamp]

Fixes JXPATH-204.

Problem

UnionContext (the EvalContext produced for a union expression like a | b)
only populates its NodeSet lazily, inside setPosition(). Normal XPath
evaluation always calls setPosition()/nextNode() before reading values, so
this is invisible in most usages.

However, ExtensionFunction.computeValue() resolves each argument by calling
getValue() on the argument's EvalContext directly, without iterating it
first:

parameters[i] = convert(args[i].compute(context));
...
private Object convert(final Object object) {
    return object instanceof EvalContext ? ((EvalContext) object).getValue() : object;
}

EvalContext.getValue() delegates to getNodeSet(), and
NodeSetContext.getNodeSet() just returns the backing field as-is (it does
not iterate). Since UnionContext never had setPosition() invoked yet at
that point, the backing NodeSet is still empty. The net effect: a custom
extension function invoked with a union expression as an argument, e.g.
my:fn(a | b), receives an empty NodeSet instead of the union of a
and b.

Fix

Override getValue() in UnionContext to force the node list to be computed
(via getContextNodeList(), which is what setPosition()-driven iteration
would have done anyway) before delegating to super.getValue(). This matches
the fix proposed by the reporter in the JIRA issue.

@Override
public Object getValue() {
    getContextNodeList();
    return super.getValue();
}

Testing

• Added ExtensionFunctionTest.testUnionOperatorArgument(), which calls a
  NodeSet-consuming extension function (test:countPointers) with a union
  of two location paths (/beans[1] | /beans[2]) and asserts both nodes are
  seen (2), where previously it saw 0.
• Ran the new test against the unmodified UnionContext first to confirm
  it reproduces the reported bug (expected: <2> but was: <0>), then applied
  the fix and confirmed it passes.
• Full suite: mvn test → 416 tests run, 0 failures, 0 errors (1 pre-existing
  skip, unrelated to this change).
• mvn checkstyle:check pmd:check → clean.
• mvn verify → success.
• Manual runtime check outside the test framework: a small standalone program
  registering a custom int countPointers(NodeSet) extension function and
  evaluating my:countPointers(/beans[1] | /beans[2]) via JXPathContext
  printed 0 (bug) against the unfixed jar and 2 (correct) against the
  fixed jar, confirming the fix holds at the public API level, not just
  inside the unit test.

Checklist (from PR template)

• Read the contribution guidelines for this project.
• Read the ASF Generative Tooling Guidance.
• AI was used to help prepare this pull request: Claude Code (Claude Sonnet 5) was used to investigate the report, write the fix and regression test, and run the build/tests described above. The diagnosis and the exact fix shape were laid out by the reporter in the JIRA issue; the AI tool implemented, tested, and verified it.
• Run a successful build using the default Maven goal (mvn).
• Write unit tests that match behavioral changes, where the tests fail if the changes to the runtime are not applied.
• Write a pull request description that is detailed enough to understand what the pull request does, how, and why.
• Each commit has a meaningful subject line and body.