[Adelaiten]

Add a statically configured list of fields excluded from the LDF output, exposed in the library and the CLI, and shorten the encoding of end edges to save tokens.

exclude_fields

• New option of json_to_lineage_format(graph, {"exclude_fields": [...]}). The default (empty list) leaves the output unchanged.
• Matching: a bare name removes the field at every nesting level; a dotted path ("technology.id") removes only that field, arrays take no indices and paths are relative to the asset/edge; removing an object or array removes its whole content; wildcards use fnmatch; unknown entries are ignored (logged at debug level).
• Exclusion runs on a copy of the input before serialization, so the output has no empty separators, entries or sections, and no trace in the header (no path=/T/TID without technology, no HP/parent index without hierarchical_path).
• Partially removed records use named forms instead of empty positional slots: chc=/cha=/chh= for children, i=/n=/t= in HP segments, n=name or bare id in BT/BC/DC/DSD, attributes without name or value, node head without type. The decoder reads all of them, so excluded output still round-trips.
• Validation: TypeError for a non-list of str; ValueError when an asset id or an edge source/target is removed.

CLI

• ldf json-to-lineage gains --exclude (comma-separated and/or repeated).
• convert_json_file_to_lineage gains an optional exclude_fields argument.

End edges

• An edge without a target is written as "N7" instead of "N7>" in E and EP. The decoder still accepts the old form and still drops such edges.
• This changes the default output for inputs that contain edges without a target (20 of 50 edges in big_query_lineage.json, 40 characters fewer). Everything else is unchanged.

Tests

• tests/test_exclude_fields.py (236 tests): synthetic graphs, validation, names, dotted paths, wildcards, no-trace checks, round trips, CLI, end edges, and an exhaustive and randomised run over big_query_lineage.json (skipped when the file is absent).

Docs: README (usage, exclude_fields table, CLI example, E section) and CHANGELOG.

[gcattan]

Ok, spot a bug with the decoder. Pushing some fix. Not related to your PR, but the type of the edge was dropped.

The token saving is less than expected, because >\n is in fact a single token.