perf: project imported matches to slim structs in standings import

[Rathoz]

Summary

• Imported matches are now projected to a StandingsImportMatch struct — {matchId, finished, winner, opponents (by ref), slim games} — at import time in Parse/Lpdb.lua. Tiebreakers (Buchholz, Game/Util, Game/Rounds/Util) are the only consumers and read exactly these fields. This drops bracketData, extradata, and full game payloads per match (big memory win on large group stages), and matchFromRecord now runs once per match2 record instead of once per (match, round).
• mergeOpponentsData in Table.lua now uses a name-keyed opponent index (O(1) fast path, one linear-scan fallback for renamed teams via Opponent.same) and an explicit allocation-safe merge that reproduces deepMerge's output without mutating imported structures. The Table module import is removed as it is no longer needed.
• Type annotations updated: TiebreakerOpponent.matches and StandingTableOpponentData.rounds.match/matches now reference StandingsImportMatch instead of MatchGroupUtilMatch.

How did you test this change?

• busted --run=ci: 605 successes / 0 failures / 0 errors (standings_import_spec, standings_table_merge_spec, standings_tiebreaker_spec all pass)
• luacheck on all three touched files: 0 warnings / 0 errors
• Based on standings-tests-ai (PR test: add integration tests for standings #7647) whose import/merge/tiebreaker specs lock the observable semantics; no test files were modified.

🤖 Generated with Claude Code

[hjpalpha]

seems reasonable on phone