Fix Reassure CI flakiness with delta thresholds and Boolean parsing

Two changes:

1. Raise delta check thresholds from 10ms/20% to 20ms/40% to match stability
   check thresholds — shared CI runners introduce ~10-15ms of jitter which
   makes the previous 10ms absolute threshold too strict.

2. Fix Boolean('false') bug in IS_VALIDATING_STABILITY parsing — the string
   'false' was coerced to true, causing delta check failures to be
   misreported as stability check failures.

Author: abzokhattab