[jsonbailey]

Summary

The reconnect delay, the backoff-reset window, and the time-limit error strategy measured elapsed time with the wall clock (time.time()). A clock change during an outage could stretch the reconnect sleep by the size of the change, or skip backoff entirely.

• Use time.monotonic() for all interval arithmetic in the sync and async clients and in _TimeLimitErrorStrategy. The retry-jitter random seed still uses time.time().
• Add tests that step the wall clock back one hour on each read and confirm the delay and time limit are unaffected.

---

Note

Overview
Reconnect timing no longer follows wall-clock jumps. Sync and async SSE clients now use time.monotonic() for reconnect sleep math, disconnect timestamps, and the backoff-reset window instead of time.time(), so NTP or manual clock changes during an outage cannot inflate reconnect delays or incorrectly reset backoff.

Time-limited error retry uses the same clock. _TimeLimitErrorStrategy measures elapsed retry time with time.monotonic() as well.

Tests sleep slightly past time limits on Windows (~15ms clock steps) and add cases that advance time.time() by an hour per read while asserting reconnect delays and backoff reset behavior stay correct.

^{Reviewed by Cursor Bugbot for commit b52853c. Bugbot is set up for automated code reviews on this repo. Configure here.}