fix(config): keep empty facet arrays as NamedList, not List

[adityamparikh]

JsonResponseParser classifies JSON arrays purely by shape: an array of
alternating [String, non-String, ...] pairs becomes a NamedList, anything
else becomes a List. An empty array has no shape to inspect, so a facet on a
field that matched zero documents is converted to an empty List.

SolrJ's QueryResponse casts every facet_counts/facet_fields entry to
NamedList, so that produces a ClassCastException whenever a faceted field
has no matches — a plausible query, not an edge case:

{"facet_counts":{"facet_fields":{"genre":[],"author":["asimov",3]}}}
                                          ^^ empty -> List -> CCE

Fix

Give the traversal positional context instead of guessing: thread the node path
through toNamedList/convertValue, and treat anything directly under
facet_counts/facet_fields as a NamedList regardless of shape. The existing
shape heuristic still covers the other flat-NamedList sites.

Tests

Adds JsonResponseParserTest (5 cases), covering populated facets, the empty
facet, empty/plain arrays outside facet_fields (which must stay List),
and an end-to-end case that feeds the parsed response into a real
QueryResponse.

Verified against main: reverting only the JsonResponseParser change while
keeping the new test fails 2 of the 5 cases, one of them with the
ClassCastException above. With the fix, all 5 pass alongside the existing
JsonResponseParserContentTypesTest.

./gradlew spotlessCheck test is green.

🤖 Generated with Claude Code

[adityamparikh]

Rebased onto main (a84033b).

One conflict, in JsonResponseParser.convertValue: main added a @Nullable return annotation (jspecify rollout) on the same signature this PR threads a path parameter through. Kept both — private @Nullable Object convertValue(JsonNode node, String path).

No other changes; the fix and its tests are as reviewed. Verified: ./gradlew build green on JDK 25.

[adityamparikh]

Added an end-to-end test and closed a duplicate PR in this one's favour.

Duplicate: I opened #185 against the same bug without spotting this PR — my duplicate check
searched issues, not pull requests. #185 is now closed, and issue #182 (which I filed for the
same defect) points here.

Comparing the two implementations, this one is better and is what should merge. It anchors on
the full path facet_counts/facet_fields/<field>, so a key named facet_fields appearing
anywhere else in a response cannot be misread. #185 matched the bare key at any depth. #185 also
covered facet_queries and facet_intervals, but that turned out to be dead weight — Solr emits
both as JSON objects, which already decode correctly through convertObject.

Ported across (e4d0646): the one thing #185 had that this PR didn't — an integration test
against real Solr via Testcontainers.

The existing JsonResponseParserTest pins decoding at the parser boundary with a hand-written
payload, which is the right place for it. But that leaves one assumption unverified: that a real
Solr genuinely emits [] for a zero-hit facet. If Solr ever emitted {} instead, those unit
tests would keep passing while the bug they guard had stopped existing in that form. The new test
closes that gap by faceting a filter designed to match nothing and asserting an empty facet map
comes back.

Verified rather than assumed: reverting only the JsonResponseParser change on this branch and
re-running makes the new test fail with java.lang.ClassCastException. With the fix it passes.
Full build is 378 tests, 0 failures, 7 skipped.