ax-normalize — any corpus → one [RecordSet]

The article's normalization promise: "given any corpus of information regardless of its format, we'll normalize it." This crate maps every recognized format onto the engine-independent [RecordSet] from ax-core, so detectors never see the difference between a CSV and a Parquet file.

Formats are plugins: each is an independent [FormatParser] (one file under [parsers]), resolved by a [ParserRegistry] via file extension then content sniff. Adding a format is a new file plus one registration line — see [parsers::default_registry]. Binary columnar formats (Parquet, Arrow IPC) live behind the default-on polars feature.

Normalization is deterministic: column order is stable (header order for tabular input, sorted key-union for JSON), and absence is explicit — a key missing from one JSON row becomes [ax_core::Value::Null], never a guess.