Start() rejected only zero and negative intervals, but Task.Delay
truncates to whole milliseconds and tops out at int.MaxValue ms on
netstandard. A sub-millisecond interval became Task.Delay(0) and spun a
core, which the zero check was added to prevent, and one over about 24.8
days ran the action once and then faulted the loop. Reject anything
outside [1 ms, int.MaxValue ms] with the existing exception.

Fixes #65

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vpup29Ycn8SGi7LvBjAq1c

… faulting

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vpup29Ycn8SGi7LvBjAq1c

…-interval-range

Reject polling intervals Task.Delay cannot honor

…[patch]

An asynchronous action that threw an OperationCanceledException of its
own, such as the TaskCanceledException HttpClient reports a timeout with,
left its run Canceled rather than Faulted. TryRun and RethrowExceptions
only looked at Task.Exception, which a Canceled task does not carry, so
the failure was cleared and never reported.

A run that Stop() cancels already ends normally, so nothing cancels a run
on purpose. A Canceled run is now treated as a failure: TryRun rethrows the
action's exception into the polling loop, and RethrowExceptions reports it.

Fixes #66

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017crwGv42t1FV1Sp9QxoQNe

Report an action that throws OperationCanceledException as a failure [patch]

TryRun claimed the ActionTask slot under the lock, but the action body ran
later on the thread pool, and Stop() only cleared ShouldPoll. A run that
had been dispatched but not begun could start after Stop() returned and
after the polling loop exited, so stopping and then disposing what the
action used was not safe. The loop also read ShouldPoll, released the
lock and called TryRun, which never looked again.

A run now carries the stop generation it was claimed under and re-checks
it under the lock before calling the action, and the polling loop passes
its generation to TryRun so a stop in that gap prevents the claim.

StopAsync() stops, then waits for any pending restart, the polling loop
and the last run, so nothing is executing and nothing can start once it
completes. The README listed a Dispose() that does not exist; it now
documents StopAsync instead.

Fixes #67

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K6bDUGMFsAQews3TnA2jXr

Add StopAsync and keep a dispatched run from starting after Stop [minor]