`Interval` scheduled rows: reschedule after duration starting from run time, not from `now` after execution finished

[gefjon]

In crates/core/src/host/scheduler.rs SchedulerActor::delete_scheduled_reducer_row, when rescheduling a reducer for a ScheduleAt::Interval, we do Timestamp::now() + dur to get the next time to run. This means that any interval reducer which has non-negligible execution time will gradually drift from its intended interval. Change this so that we reschedule for prev_scheduled_time + dur instead.

[gefjon]

We've had numerous complaints about scheduler drift in the public Discord, which I believe are due to this.

[gefjon]

When re-scheduling a reducer that takes longer to execute than its requested Interval, log a warning (preferably to the module's logs, but realistically to the host logs given that we don't have a good way to do that right now) and re-schedule for prev_scheduled_time + (n * dur), where n is the smallest integer such that prev_scheduled_time + (n * dur) is greater than Timestamp::now().

[Shubham8287]

I think, this other issue also tracks same - #2648

[Shubham8287]

Not similar, but related - #2884.