fix(trpc): let the hourly MRR refresh collect the run it started (#6931)

refreshSigmaMrr ignored the cache only on its first step, the one that
starts a Sigma run. Every poll after that honoured the cache, found the
previous 12h entry still valid, and returned it — so the job reported
refreshed while the run it had just started was never collected and its
pending handle expired. The admin MRR tile only moved when the entry
lapsed, up to 12h behind what Sigma had already computed, and each
hourly run was burnt for nothing (~24 a day).

Every step now ignores the cache; the finished run is returned straight
from the pending branch, which is where it always landed.

Author: saddlepaddle

packages/trpc/src/router/analytics/business.ts

@@ -270,18 +270,19 @@ export async function refreshSigmaMrr({
 	pollIntervalMs?: number;
 } = {}): Promise<MrrResult> {
 	const startedAt = Date.now();
-	// Ignore the cache on the first step only. Otherwise a scheduled refresh
-	// that runs more often than the entry expires would read its own previous
-	// result and never actually refresh. Subsequent steps must consult the
-	// cache, since that is where the finished run lands.
+	// Ignore the cache on every step. The job runs more often than the entry
+	// expires, so honouring it would hand back the previous result before the
+	// pending run is ever collected — the tile would then only move when the
+	// entry lapsed, with a Sigma run burnt every hour for nothing. The finished
+	// run is returned straight from the pending branch, not via the cache.
 	let last = await advanceSigmaMrr({ ignoreCache: true });
 	while (
 		!last.available &&
 		last.reason === MRR_COMPUTING_REASON &&
 		Date.now() - startedAt < timeBudgetMs
 	) {
 		await new Promise((resolve) => setTimeout(resolve, pollIntervalMs));
-		last = await advanceSigmaMrr();
+		last = await advanceSigmaMrr({ ignoreCache: true });
 	}
 	return last;
 }