fix(parsers): stop Ruby nested classes deleting their parent, and extract method parameters - #1581

Ten PRs since 0.5.6 — mostly parser correctness, all with regression tests.

Parsers
- Ruby nested classes no longer delete their parent; method parameters
  extracted at all (#1523, #1527, #1581)
- C/C++ functions returning a pointer or reference are extracted, and
  registered in pre_scan so they resolve as call targets (#1524, #1582)
- C# fields and locals parsed, restoring receiver-type inference for
  `var x = new T(); x.M();` (#1525, #1584)
- JavaScript destructured, array and rest parameters extracted, recording
  the binding names rather than a placeholder (#1527, #1591)

Queries and analysis
- Five defects that returned wrong answers without erroring: inflated
  repository stats, misattributed callers, discarded graph_name,
  count(*) after OPTIONAL MATCH, an ignored tool argument (#1577)
- Dead-code detection made correct and self-consistent, including the
  JS module-level false positive (#600, #1559, #1578)

Indexing
- add_package_to_graph on a flat module no longer indexes the whole
  virtualenv (#1528, #1586)
- SCIP job progress no longer exceeds 100% (#1535, #1583)

Docs and tests
- Database Options table columns realigned (#1590)
- MCP tool definition contract validation (#1580)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

…issues; finish the #1527 parameter audit (#1648)

* perf(kuzu): batch relationship-only UNWIND MERGEs; scope the per-row fallback to node-MERGE shapes (#1605)

Kuzu 0.11.3's MERGE pipeline mis-binds when a merged NODE's key repeats
non-adjacently in one UNWIND batch ([(A,P),(B,X),(C,P)] binds C to X) —
reproduced minimally and pinned in tests. Relationship-only MERGEs
(MATCH..MATCH..MERGE rel) do not exhibit the bug, verified with interleaved
duplicate endpoint keys and duplicate pairs, so they now run batched: the
full integration suite drops from ~5:06 to ~4:09 (-20%) and a sample-project
index halves. Inheritance batches are also deduped writer-side so a repeated
(child, parent) record cannot double an INHERITS edge on embedded backends.

* fix(indexer): retry a failed per-file graph write once before recording it (#1612)

A transient write failure under runner load silently cost the graph a whole
file's edges (LadybugDB intermittently 51 CONTAINS short in the parity run).
Writes are MERGE-idempotent, so one retry converts the blip into success;
persistent failures still land in write_failures and the error log. Together
with the #1605 batching change (which removes most of the per-row write
pressure the flake correlated with), this addresses the observed drop.

* test(cgcignore): make the CLI cases hermetic and move them to the integration suite (#1422)

The TC-12+ cases shared one fixed /tmp/cgc_test/_home database across every
test and run, wiped it mid-suite, and were gated on a live Neo4j — so they
were permanently skipped on most machines and intermittently failed with a
different name each full-suite run wherever they did run. Each test now gets
a fresh HOME and an explicit embedded backend (no server, no gate, no wipe),
and the subprocess-driven cases live in tests/integration where that cost
belongs. tc01-tc11 (pure PathSpec logic) stay in the unit suite.

* fix(cpp): extract lambda parameters through the shared param helper (#1527)

The inline version first clobbered its correctly-walked parameter_list with
a nonexistent 'parameters' field on lambda_expression, then filtered for
bare identifiers over parameter_declaration nodes — every lambda assignment
got args: []. lambda_expression carries the same declarator→parameters field
chain as function_definition, so the shared extractor handles it, rendering
args in the project's established 'type name' form. Last remaining case of
the #1527 parameter audit (Ruby #1581, PHP key + #1643, JS #1591 all landed
previously).

* regen cpp golden: lambda Parameter nodes now extracted