Support bzip2 and xz compressed inputs in LLM file analysis

[guan404ming]

Why

• LLMFileAnalysisOperator only accepted gzip, so archived logs and exports stored as .log.xz or .csv.bz2 were rejected outright.
• The path suffixes were already recognised, but any codec other than gzip was refused, forcing users to decompress in a separate task.

How

• Accept bzip2 and xz alongside gzip for .log, .json, and .csv inputs, using the standard library only.
• Replace the hardcoded gzip branch with a codec lookup, so all supported codecs share one read path.
• Keep snappy and zstd rejected, and keep compression unsupported for Parquet, Avro, image, and PDF inputs.

---

Was generative AI tooling used to co-author this PR?

• Yes — Claude Code (Opus 4.8)

Generated-by: Claude Code (Opus 4.8) following the guidelines