thread - Replace O(n^2) origin-id/stanza-id lookups in AppModel with hash indexes and stop rebuilding the whole index on every unseen stanza-id, so archive application no longer grows quadratically with history size. - Run OMEMO decryption on a serial background queue instead of the main actor, and decrypt archived stanzas in small batches with a shorter yield, keeping the UI responsive during large archive catch-ups. - Show the "Синхронизация истории…" banner only for the one-page bootstrap window; incremental backlog catch-up continues silently in the background. - Fix MUC-MAM: publish decoded group mutations at the end of room catch-up; they were previously accumulated and then silently dropped. - Retry a failed catch-up pass automatically (bounded) and lengthen the query/apply timeouts so a single slow page no longer leaves history unloaded until the next app activation.