pub fn train_dict(sorted_unique: &[String]) -> Option<Vec<u8>>Expand description
train ONE deterministic zstd dictionary over the canonical texts.
the samples are sorted and deduplicated by the caller (the writer) so the
trainer input is canonical; from_continuous is then a pure function of
the concatenated bytes + per-sample sizes, pinned to the workspace zstd
version. returns None when there is too little material to train a
useful dictionary (the trainer needs a handful of samples); the caller
then simply does not offer the dict variant for that build.