fix(scheduler): reject zero/negative interval in every_N cooldown parsing

[shaun0927]

Problem

After PR #76 (which fixed the ValueError/IndexError crashes from
malformed every_* repeat values), _parse_cooldown('every_0h') still
silently returns timedelta(0):

>>> _parse_cooldown('every_0h')
datetime.timedelta(0)

Inside check(), the cooldown gate is:

last = _last_run(tid, done_files)
cooldown = _parse_cooldown(repeat)
if last and (now - last) < cooldown: continue

With cooldown = timedelta(0), (now - last) < cooldown is False from the
first cycle onward, so the task fires on every 120-second scheduler tick
instead of every N hours. For a side-effectful prompt (e.g. anything that
posts a message, schedules another task, or triggers a code_run with
external calls) this is a runaway loop.

Fix

Add a positive-interval guard inside the existing try block so the
malformed value falls through to the same warning + 20h fallback that
every_h and every_xyz already use:

     if repeat.startswith('every_'):
         try:
             parts = repeat.split('_')
             n = int(parts[1].rstrip('hdm'))
+            if n <= 0: raise ValueError(f'positive interval required, got {n}')
             u = parts[1][-1]
             if u == 'h': return timedelta(hours=n)
             if u == 'm': return timedelta(minutes=n)
             if u == 'd': return timedelta(days=n)
         except (ValueError, IndexError):
             pass  # fall through to warning below

This keeps the change radius local to _parse_cooldown and reuses PR #76's
existing fallback path; no new control flow is added.

Verification

REPL on the patched function:

every_0h  -> 20:00:00            (fallback)
every_-1h -> 20:00:00            (fallback, ValueError raised by int())
every_h   -> 20:00:00            (fallback)
every_1h  -> 1:00:00
every_30m -> 0:30:00
once      -> 999999 days, 0:00:00

Builds on PR #76.