v0.1.0

Initial public release.

Added

• head N / tail N — first/last N records, streaming; tail keeps only an
  N-record ring buffer.
• select FIELDS — project comma-separated dotted paths (id,user.name,ts).
• filter EXPR — keep records matching a safe no-eval predicate grammar:
  ==,!=,<,<=,>,>=, exists, contains, and/or/not, parentheses,
  dotted paths.
• stats — per-field presence %, null %, observed types, numeric
  min/max/mean/stddev, string cardinality; nested keys flattened to dotted paths.
• schema — infer a JSON Schema (draft 2020-12, structural subset) over a full pass.
• validate — report malformed lines by 1-indexed number; with --schema,
  also report non-conforming records. Exits non-zero on any failure (CI-friendly).
• Global --limit N — stop after N input records.
• Transparent .gz input; reads a file arg or stdin.
• Requires Python 3.12+ (argparse in 3.12 handles options interspersed with
  positional arguments, e.g. jlkit select name --limit 1 data.jsonl).

Changed

• A missing or unreadable input file now prints a one-line
  jlkit CMD: cannot open '...': <reason> (exit 2) instead of a traceback.
• Writing to a closed pipe (jlkit ... | head) exits quietly instead of
  raising BrokenPipeError on shutdown.