Contract 6.6.0

Deck archetypes, phase 2 (design §6, §7, §12.2): the archetype is stamped on every deck, so a whole population can be folded or filtered by name, and the fit reader says which shapes a player already fields.

• battles_meta_decks { group_by: "archetype" | "family" }: the population's decks folded by stamped label ("Royal Hogs bridge spam") or by family, one row each with decks, battles, wins, losses, win_rate, players and share; on a clan, player or collection segment each row carries members[] - who plays the shape, with battles, wins and their most-played deck of it - and players is exact. Sorted by players then battles: who plays what, never a tier list (shrunk_win_rate is deliberately absent). decks[] is empty with group_by. This is Elixir Clan's "what decks do our players use?" in one call.
• The archetype filter now runs over the stamp for every deck over min_battles (the 2,000-row bound of 6.5.0 is gone).
• fit_for on battles_meta_decks: fit_for.plays lists the families and labels the player fielded in the window, and every row's fit carries plays_family and plays_archetype, with a note ranking adoption cost - a row in a family they play costs the least, the same family with a different win condition is the usual next step, a new family is a new deck to learn as well as levels to buy.
• The stamp (migration 0148: deck.archetype_family / archetype_label / archetype_win_conditions / archetype_version) is written at deck insert and caught up nightly for every row behind the current grammar + vocabulary version; {archetype_stamp} on the migrate Lambda runs the same on demand. A vocabulary import or a rule change reaches history by the next morning.

Additive.