fix: retry hash clients when timeouts are reached - #655

HashClient waits until elapsed time is strictly greater than its retry/dead timeout. With a zero timeout, a second call in the same clock tick therefore skips the retry or reports that all servers are down. This also makes the existing immediate-retry tests depend on clock resolution.

Treat the timeout as elapsed when the interval equals the configured value, consistently for ordinary commands, set_many, and dead-server recovery. Freeze the clock in the existing zero-timeout tests and add boundary coverage for zero and positive retry timeouts. The positive-timeout cases also verify that calls before the deadline do not retry.

Validation: two existing tests with a frozen clock and four new boundary cases fail before the change. The unit run after the fix has 496 passed, one skipped, and one Windows-only failure because this Python build lacks AF_UNIX. Black 24.4.2, Flake8 7.1.1, and diff whitespace checks pass. Compression tests were excluded because zstd could not be built locally; live memcached integration tests were not run.