refactor(crawler): tighten FessCrawlerThread hot path

[marevol]

Summary

Low-risk, implementation-level cleanups on the per-URL hot path in
FessCrawlerThread. No architectural or interface changes, and no intended
behavior changes.

Scope is intentionally narrow to minimize regression risk. Larger structural
ideas (e.g. batching IndexingHelper.getDocument via idsQuery) were
considered but deferred to separate PRs after profiling.

Changes

src/main/java/org/codelibs/fess/crawler/FessCrawlerThread.java:

• Deduplicate ComponentUtil.getFessConfig() in isContentUpdated. The
  config was fetched twice (once for isIncrementalCrawling(), then again a
  few lines later). Hoisted to a single call at the top of the method.
• Replace stream(getPermissions()).of(s -> s.forEach(list::add)) with a
  null-guarded Collections.addAll. Drops the Stream pipeline and lambda
  allocation per call while preserving the original null-array tolerance of
  StreamUtil.stream.
• Drop the redundant blank-URL filter in storeChildUrlsToQueue.
  getAnchorSet / getChildUrlSet (upstream) and CrawlerThread.storeChildUrls
  (downstream) already drop blank URLs, so the intermediate
  stream().filter().collect(toSet()) was wasted work and an extra Set
  allocation per call.
• Guard three logger.debug sites with isDebugEnabled() short-circuits
  (L168 / L195 / L217). L217 in particular evaluated
  fessConfig.getIndexFieldExpires() for the format arg unconditionally; the
  other two avoid Log4j2 LogEvent setup on the cold path.
• Cache responseData.getLastModified() in a local variable. It was
  called twice (null check, then .getTime()); now once.
• Fix a latent NPE in getClientRuleList. When a rule string is
  malformed (no :), the computeIfAbsent mapping function returns null.
  ConcurrentHashMap does not cache null values, but Stream.map does
  emit null into the resulting List, and getClient later dereferences
  Pair.getSecond(). Added .filter(Objects::nonNull) before .toList().
• Cleanup: drop unused StreamUtil.stream / Collectors imports,
  remove a stale // add an url comment, and reword the
  storeChildUrlsToQueue Javadoc to match the new behavior.

Intentionally NOT changed

• LinkedHashSet in getAnchorSet — preserving anchor insertion order is
  visible behavior for crawl prioritization.
• The two HEAD-request paths in isContentUpdated are already mutually
  exclusive via the if (responseData == null) guard; no actual duplicate
  request occurs.
• Batching IndexingHelper.getDocument via idsQuery — requires changes to
  the crawler main loop and is out of scope for a low-risk refactor.

Test plan

• mvn compile — BUILD SUCCESS
• mvn test -Dtest=FessCrawlerThreadTest — 14 tests pass, 0 failures
• mvn formatter:format && mvn license:format — applied
• Manual crawl smoke test against a small web/file source (recommended
  before merge)

🤖 Generated with Claude Code