Closes #485

Problem

Long setTimeout delays drift (OS sleep, GC, throttling, clock skew). On daily/weekly schedules a heartbeat could wake a few seconds late, fail the exact-second match(), skip the execution and emit a spurious missed execution warning. The miss rate scaled with timer duration, so daily/weekly jobs were hit hardest.

Reported in #485 (with a strong inspector-level diagnosis showing the timer fires but the callback is skipped, miss rate correlating with timer duration) and the WSL2 clock-drift reports in #475.

Fix

New configurable option missedExecutionTolerance (ms, default 1000): a heartbeat that wakes within the tolerance of the slot it was armed for runs that slot instead of reporting it missed.

The tolerance is always bounded by the gap to the next slot, so:

• a late run can never bleed into the following slot, and
• a tolerance larger than the cron interval can never run a slot twice.

The run / miss / wait decision is extracted into a pure planBeat function with exhaustive scenario tests (on-time, late-within-tolerance, late-beyond-tolerance, early fire, the 1s-cron cap, hourly recovery, the superseded-slot boundary, long-block catch-up). The runner now drives off the slot it armed for rather than re-matching the late wall clock.

The existing "missed execution" suppression (suppressMissedWarning and the auto-suppress when an execution:missed listener is attached) is unchanged.

Notes for #475 (WSL2)

That one is environmental: the WSL2 guest clock desyncs from the Windows host (microsoft/WSL#10006). The grace window softens minor drift, but a large clock jump is still a genuine miss. The confirmed user fix is an NTP/hwclock resync.