Fix possible race conditions when suspending/unsuspending accounts

[ClearlyClaire]

Currently, when suspending or unsuspending an account, the suspended_at flag is immediately updated, and the heavy work is performed in a background job. However, the background job sets the flag again (losing suspension origin in the process!) if it isn't set, which can lead to surprising and unwanted results: if an account is unsuspended before the suspension worker runs:

• in the best case scenario, the account will be suspended then promptly unsuspended by the unsuspension worker, the long-time result is not an issue, but there will be extra work and a time window in which the account will be marked as suspended for no good reason
• in the case where the suspension and unsuspension worker are picked up by different processes at the same time, the result is undefined
• in cases where the suspension or unsuspension workers fail for some reason, they will re-set the flag at each attempt, causing a highly confusing behavior (account re-suspended even though admins unsuspended the account, multiple times)
• it can change a remote suspension flag into a local suspension flag, which is very confusing

Therefore, this PR changes the suspension/unsuspension worker logics to not set the suspension flag, but check it instead: the appropriate worker will only be scheduled after the flag has been set appropriately on the call site, so this should be safe.