[vvoytovych]

Related to #978.

Every bucket comes back an hour after its last reconcile. After a restart the manager reconciles the buckets as it engages their projects (656 projects in 7 min on 2026-11-03), so their rechecks come back an hour later in the same wave, and every hour after unless an event moves one. Each recheck now waits wait.Jitter(time.Hour, 0.1): between 60 and 66 minutes, so the wave widens instead of repeating.

Corner cases

case	what happens
A bucket with no events	It comes back between 60 and 66 minutes after its last reconcile
A bucket with an event	The event reconciles it at once, as before
The buckets a restart reconciles together	Their rechecks spread over up to 6 more minutes each hour

Test plan

• TestAllowanceBucketComesBackAfterTheRecheckInterval reconciles 20 times: each recheck lies in [1 h, 1.1 h), and they are not all equal. On main all 20 are 1h0m0s, so it fails; here it passes.
• A planted factor of 0.2 turns it red. go test ./internal/quota/... passes with envtest 1.37.0.

[ecv]

Approve.

wait.Jitter(bucketRecheckInterval, 0.1) returns between 60 and 66 minutes, so buckets reconciled together after a restart stop coming back in the same minute; this is the jitter follow-up from #978.

• Jitter only ever lengthens the interval, so no bucket is rechecked sooner than before, and the stranded-reservation recovery still runs within 66 minutes.
• The test bounds every result to [1h, 1.1h) and requires more than one distinct value across 20 reconciles, so it fails if the jitter is dropped. With continuous random durations a false failure is effectively impossible.