fix(extractor): defend against archive bomb, recursion, and Zip Slip

[marevol]

Summary

Hardens ZipExtractor, TarExtractor and LhaExtractor against
content-driven attacks. The input stream is treated as untrusted; the
extractor params map remains admin-configured / trusted.

Protections added

• Recursion-depth limit — new extractorDepth param tracked through
  AbstractExtractor (EXTRACTOR_DEPTH_KEY, getCurrentDepth,
  incrementDepth, checkDepth). Each archive extractor calls
  checkDepth on entry and incrementDepth when delegating to a nested
  extractor. Default maxArchiveDepth = 10.
• Total uncompressed-byte cap (maxBytes, default 2 GiB) measured
  from the bytes actually read out of each entry stream — not from the
  declared entry.getSize() which can be -1 or fabricated.
• Maximum entry count (maxEntries, default 100,000) to defend
  against many-zero-byte-entries DoS.
• Compression-ratio threshold (maxCompressionRatio, default 100:1
  for entries > 1 MiB) using the central-directory compressed size where
  present and the bytes consumed from a CountingInputStream when the
  local header carries a data descriptor.
• Zip Slip path-traversal rejection — entry names are normalised via
  Path.normalize() and rejected when they escape the conceptual
  extraction root, are absolute, or contain .. segments.
• Tar symlink / hardlink skipping — entries with LF_SYMLINK /
  LF_LINK are skipped with a debug log so they cannot leak content
  from outside the archive.
• Configurable filename encoding (filenameEncoding, default
  UTF-8; supports CP932 / MS932 for Japanese archives).
• Early ZIP signature validation so a clearly non-ZIP blob is
  reported as ExtractException instead of returning empty content
  silently.

How to override

Field setters via DI (these are admin-configured, not params):

Setter	Default
setMaxArchiveDepth(int) (AbstractExtractor)	10
setMaxBytes(long)	2 GiB
setMaxEntries(int)	100_000
setMaxCompressionRatio(long) (Zip)	100
setFilenameEncoding(String) (Zip)	UTF-8

Param key (caller-controlled, used for recursive calls):

• extractorDepth — current depth value, parsed as integer.

Threat model

• Untrusted: archive content stream
• Trusted: extractor params map (admin-configured / internal)

Test plan

• AbstractExtractorTest: depth helper coverage (null/blank/garbage,
  incrementDepth returns NEW map, checkDepth at/above limit).
• ArchiveExtractorSecurityTest (new): 15 synthetic-archive tests:
  • test_zipBomb_byteLimit
  • test_zipBomb_entryLimit
  • test_zipSlip_pathTraversal, test_zipSlip_absolutePath
  • test_recursionDepth_exceeded,
    test_recursionDepth_belowLimit_succeeds
  • test_cp932Filename
  • test_tar_symlinkSkipped, test_tar_hardlinkSkipped,
    test_tar_pathTraversal
  • test_compressionRatioExceeded
  • test_tarBomb_byteLimit, test_tarBomb_entryLimit,
    test_tar_recursionDepth_exceeded
  • test_lha_recursionDepth_exceeded
• All previously-passing archive tests still pass
  (ZipExtractorTest, TarExtractorTest, LhaExtractorTest,
  ArchiveExtractorErrorHandlingTest).
• No regressions in the wider test suite — only env-dependent
  failures remain (Docker for SMB / S3 / GCS / Storage clients,
  LibreOffice for JodExtractor).