Run a Decision Bracket round and get a result

Bracket is a swipe-to-keep/reject flow in the browser (fetch real candidates for one category, decide each one independently, tie-break if more than one is kept) — genuinely multi-step, unlike /spin's single weighted pick, which is why Stage 42 originally left it out of this surface. This endpoint supports two shapes, chosen by whether decisions is supplied: omit it for the default "auto" mode, where the server fetches real candidates and runs the keep/reject pass itself (biased toward more-trending/closer/boosted places — or higher-rated for a movie-sourced category — not a coin flip), auto-resolving any tie with the same real data the interactive tie-break uses; or supply decisions — an array of { id, keep } referencing candidate ids returned by a prior call to this same endpoint with identical categoryMode/optionId/lat/lng — to control the outcome yourself. This endpoint is otherwise stateless (no session is stored server-side): a same-parameters follow-up call typically returns the same candidate set thanks to the shared api_cache layer, but this isn't guaranteed, so a decisions id that no longer matches a fetched candidate is simply ignored (treated as not kept). Every response includes the full per-candidate rating/trendingScore/distance/sponsored/kept breakdown, not just a bare winner — Stage 153: place-sourced candidates no longer carry a real Google rating (see /search's description), so trendingScore (this app's own free first-party selection/liked-history signal) is what actually drives their ranking and tie-break now. Requires the bracket scope. Costs 4 units of the key's rate-limit budget per call — the heaviest of the six v1 endpoints, since it fetches and scores a full candidate list rather than narrowing to one pick.