refactor(api,crawler,llm): precompile regex patterns and tighten hot paths

[marevol]

Summary

Cross-cutting cleanup across api/, crawler/, and llm/ packages: hoist repeated regex calls into static Pattern constants, make a couple of crawler caches thread-safe, and migrate the Date branch of the JSON encoder from SimpleDateFormat to java.time.format.DateTimeFormatter. No behavior changes intended.

Changes Made

API layer (org.codelibs.fess.api)

• BaseApiManager: hoist "/+" to a static Pattern and replace the long if/else chain in getFormatType with FormatType.valueOf(...) plus an IllegalArgumentException fallback to OTHER.
• WebApiManagerFactory: grow the webApiManagers array via Arrays.copyOf instead of building an ArrayList per call.
• WebApiRequest.getServletPath: cache the getQueryString() result in a local and switch the substring check to String.contains.
• SearchEngineApiManager: pre-compile "\\.\\.+" and "/+" patterns; replace the chained extension if/else with a Map.ofEntries lookup keyed by lowercase extension.
• SearchApiManager:
  • Pre-compile "/+", the JSONP callback sanitizer, and a shared ISO-8601 DateTimeFormatter.
  • Use the already-resolved query for getRelatedQueries/getRelatedContents instead of re-reading params.getQuery().
  • Use ArrayUtils.contains for the docId scan and a HashSet for favorite URL lookups (O(1) contains).
  • Migrate the Date branch of escapeJson to DateTimeFormatter with ZoneId.systemDefault().
• ChatApiManager: drop the redundant // Default constructor comment.

Crawler (org.codelibs.fess.crawler)

• FessCrawlerThread: remove local HTTP_STATUS_OK/HTTP_STATUS_NOT_FOUND constants and reuse Constants.OK_STATUS_CODE, Constants.NOT_FOUND_STATUS_CODE, and Constants.NOT_MODIFIED_STATUS.
• FessIntervalController: extract a runQuietly(description, action) helper so each delay stage shares the same swallow-and-debug-log treatment.
• AbstractFessFileTransformer: pre-compile cache/whitespace, trailing-slash, and SMB-prefix regexes; reuse the resolved config-parameter map (it cannot be null at that point because it's already used to construct a HashMap).
• FessXpathTransformer:
  • Switch fieldPrunedRuleMap and prunedTagsCache to ConcurrentHashMap and use computeIfAbsent, addressing a latent race when crawler threads populate the caches in parallel.
  • Cache compiled Pattern instances for convertUrlMap entries.
  • Extract isRobotsTagsIgnored and applyRobotsDirective so the meta <robots> and X-Robots-Tag paths no longer duplicate the same parsing block.

LLM (org.codelibs.fess.llm)

• AbstractLlmClient: pre-compile the HTML-tag and control-whitespace regexes; short-circuit wrapUserInput when the message contains no </user_input> to skip the unnecessary String.replace call.

Testing

• mvn formatter:format && mvn license:format should be re-run before merge if any conventions changed.
• New SearchApiManagerTest cases:
  • test_escapeJson_dateMatchesLegacySimpleDateFormat cross-checks the new formatter against SimpleDateFormat(CoreLibConstants.DATE_FORMAT_ISO_8601_EXTEND) across UTC, Asia/Tokyo, America/Los_Angeles, and Europe/Paris for several epoch values, guarding against regression in the SimpleDateFormat → DateTimeFormatter migration.
  • test_escapeJson_dateFormatShape asserts the output stays a quoted yyyy-MM-dd'T'HH:mm:ss.SSS±HHmm string.
• Existing crawler/API tests should keep passing — none of these changes alter visible behavior.

Breaking Changes

None.

Additional Notes

• The ConcurrentHashMap switch in FessXpathTransformer is the only change with potential runtime behavior implications — it's intentionally there because the caches are populated lazily from multiple crawler threads, and previously HashMap.put could race with reads.
• escapeJson(Date) now uses ZoneId.systemDefault() to mirror the previous SimpleDateFormat (which also used the default JVM zone). The new tests pin this behavior.
• Reviewers may want to spot-check the FormatType.valueOf fallback in BaseApiManager#getFormatType — the previous chain rejected unknown values via the final default arm, the new code does the same via the catch.