Team-game ladder support: opponents lists, opposing-team head-to-head, live metadata shape

[amrtgaber]

Summary

Implements #292 — team boards become supported tournament ladders — and fixes two real integration bugs found by verifying every assumption against the live upstream API first (payloads fetched 2026-07-25; test players from docs/data-sources.md).

Ground-truth findings (drove the design)

1. Upstream schema drift, latent since 2026-06-01: leaderboard entries carry their matchtype ids in leaderboardmap: [{matchtype_id, statgroup_type, ...}] — the matchtypes key the parser read no longer exists. That silent key move was the real cause of the June 1 empty-map incident; since then the derived map has been empty every cycle and only the {6: 3} floor kept 1v1 RM matches attributed — every other board's matches have been written with leaderboard_id = NULL. The parser now reads leaderboardmap with the legacy key as fallback (_leaderboard_matchtype_ids).
2. The Reject non-1v1 leaderboards at tournament create/edit #302 guard's name check was wrong against live data: real leaderboard names use the SOLO_*/TEAM_* taxonomy (SOLO_RM_RANKED), not "1v1 …" — the token check would have 422'd every create in production. Fixtures had drifted from the payload. (No live impact: prod is dormant.) Replaced by an id allowlist, and fixtures across the suite now mirror the real payload shape.
3. All team boards are statgroup_type: 1 — AoE2 team ladders rate players individually with solo statgroups, so the existing personal-stat parsing is already correct for TG; no poller stat changes needed. A live 3v3 match confirmed per-player teamid/outcome/old|newrating, all of which the match_players model already stores for every participant.

Changes

• Guard (SUPPORTED_TOURNAMENT_LEADERBOARD_IDS = {1, 2, 3, 4, 13, 14}): the six classic ranked solo + team boards (DM/RM/EW), transcribed from the live payload. Excluded until verified: SOLO_BR (FFA breaks two-team reads), CO_*, POM/RBW customs. Create/PATCH message updated.
• Matchtype floor extended from {6: 3} to all seven public ranked boards' mappings (transcribed, not guessed), with the corrected incident history in the comment.
• Opponents as lists (RecentMatchup.opponents: list[MatchupOpponent]): the old limit(1) correlated subqueries — which silently reported one arbitrary opponent on team boards — are replaced by one batched participants query over the capped matchups. The legacy scalar opponent_* fields keep their exact pre-Team-game ladder support: multi-opponent matchups, full statgroups, verified matchtype floor #292 contract: populated iff exactly one opponent (1v1), null otherwise — 1v1 consumers are byte-identical.
• Head-to-head on team boards: candidates now require rostered entrants on ≥2 distinct teams (allies queuing together is not a head-to-head; equivalent condition on 1v1), and each entrant carries team_id (additive) so the consumer can group sides. Only rostered entrants are listed, as before.
• docs/data-sources.md records the schema drift + taxonomy so the next integration reads the current truth.

Tests

588 passed. New: leaderboardmap + legacy parser shapes; extended-floor merge behavior; guard allow/deny matrix (team board 201, BR 422, unranked 422, unknown 422, EW 201); TG matchup opponents list with allies excluded + scalar nulling; 1v1 scalar byte-compatibility; allies-vs-opposed head-to-head with team_id. Consumers regenerate orval clients (additive fields only).

Closes #292 (and supersedes the #302 name check).