Setting tolerance to zero milliseconds causes strange behavior in scheduling

Steps to reproduce: Schedule a periodic task. When the tolerance is different from zero, everything works as expected. When tolerance is set to zero, the task (every now and then) will experience random deviations from the scheduled time.

Code (blinky example):

```
#include "mbed-drivers/mbed.h"
#include "mbed-hal/lp_ticker_api.h"
using minar::Scheduler;

DigitalOut led(LED1);

static void blinky(void) {
led = !led;
printf("%lu \r\n", lp_ticker_read());
}

void app_start(int, char**){
Scheduler::postCallback(blinky).tolerance(minar::milliseconds(0)).period(minar::milliseconds(1000));
}
```

I was using an FRDM-K64F board

Start with the blinky example in the issue and reproduce the timing deviations on an FRDM-K64F with tolerance set to zero, comparing them with a nonzero tolerance. Trace Scheduler::postCallback, tolerance, period, and lp_ticker_read to locate where the schedule diverges. Done means periodic callbacks remain aligned with the scheduled time when zero tolerance is used.

Written by the indexing model from the issue text.