betting_combat.rounds.features

Strategy 4 (rounds) feature library: the round-start datasets from raw UFC tables.

Port of the research ufc/dataset/ (sources, fighters, matchup, context, similarity, models, rounds, targets, build) with parity to its frozen outputs (tests/parity/test_rounds_features.py). Pure: every input arrives as an argument (UfcTables); nothing here reads a file or knows a research path.

UfcTables -> RawData -> RoundDataset(raw).build(model_features=..., live_markets=...)
    FighterHistory  group 1   as-of fighter profiles
    Matchup         group 2   A - B / A + B and style interactions
    CardContext     context   card segment, slot, division, event, venue
    StyleSimilarity group 2   style clusters and style-vs-style records
    Targets         targets   round rows and outcomes
    RoundStats      group 3   rounds fought so far (dataset 2)
    ClockFeatures   group 5   walk-forward p0 model and Bayesian finish-time clock