…licationReadyEvent listener

GraphBootstrapper's H2->Neo4j fallback was registered for
@eventlistener(ApplicationReadyEvent.class), but that event never
fires during a normal serve run because ServeCommand.call() blocks
in Thread.join() as a CommandLineRunner. Consequence: if someone
shipped a bundle with only the H2 analysis cache (no enriched Neo4j
graph) and ran `code-iq serve`, Neo4j would stay empty forever and
graph-traversal queries would return nothing. Not observed in our
pipeline today only because `enrich` always runs before `serve`.

Fix:
- Drop @eventlistener from bootstrapNeo4jFromCache() and its
  associated imports (ApplicationReadyEvent, @eventlistener).
- ServeCommand.call() now calls graphBootstrapper.bootstrapNeo4jFromCache()
  explicitly before the Neo4j status report, so the advertised
  node/edge counts reflect post-bootstrap state. Injection is
  Optional since the bean only exists in the "serving" profile
  (matches the existing @Profile + @ConditionalOnBean guards).
- Class javadoc updated to document the new invocation model and
  the reason the event-listener path did not work.

The method is idempotent (guards on count>0 and on missing H2 file),
so moving the invocation is safe even in scenarios where Neo4j is
already populated by `enrich`.

Tests: existing GraphBootstrapperTest still passes unchanged (it calls
the method directly).

Integration verification:
  # full pipeline to populate both H2 + Neo4j
  ./scripts/baseline/run-pipeline.sh spring-petclinic
  # wipe ONLY Neo4j; keep H2
  rm -rf .seeds/spring-petclinic/.code-iq/graph
  # re-run serve
  java -jar target/code-iq-*-cli.jar serve .seeds/spring-petclinic --port 18080 &
  # wait for /actuator/health=200, then:
  curl http://127.0.0.1:18080/api/stats

Before this fix: Neo4j empty, stats=0.
After this fix:  log shows "Bootstrapped Neo4j with 634 nodes and
                  604 edges from H2 cache", /api/stats returns 677
                  nodes / 604 edges / 65 files.