runProbeLoop used to sleep the full jittered interval after every
probe, so a 100ms --interval with a 30ms probe actually produced a
130ms period — the flag quietly lied about cadence and the effective
rate depended on backend latency. The fix subtracts result.Duration
from the sleep so the loop's period matches --interval. Probes that
overrun the interval clamp the sleep to zero and fire immediately
without trying to catch up on missed cycles, so a slow backend
doesn't get flooded with back-to-back retries exactly when it's
already struggling.
Pim van Pelt
695ebc4bd1