Add uwsgi-style process manager with heartbeat watchdog

[eyalr]

Summary

Adds a robust process supervision layer to the PySOA standalone server, modeled after uwsgi's worker management approach. The parent process now monitors child workers via a shared-memory heartbeat and automatically kills any worker that hangs mid-request, then respawns a replacement.

What changed

standalone.py — process manager overhaul

• Changed the default --fork-processes from 0 (no forking) to 1 (always fork at least one worker). Use --fork 0 to run directly in the current process.
• New CLI arguments:
  • --ping-timeout (default 10 s): per-request watchdog; kills a child that hasn't pinged within this window.
  • --startup-timeout (default 60 s): how long a freshly-spawned child has to send its first ping before being killed.
  • --process-shutdown-timeout (default 30 s): grace period after SIGTERM before escalating to SIGKILL on shutdown.
• _ProcessMonitor now uses a short-polling loop (join(timeout=1s)) instead of a blocking join(), enabling timely reaction to shutdown signals and hung-worker detection.
• Added _kill_process() helper for SIGKILL with logging and a final join().
• Ping state (timestamp + process start time) is reset in _start_process() before every (re)spawn to prevent a stale timestamp from a crashed child from instantly killing the replacement.

server.py — child-side heartbeat

• Added _ping_timestamp (a multiprocessing.Value('d') supplied by the parent) and _ping_parent() which writes time.monotonic() into it.
• handle_next_request() calls _ping_parent() at three points: on MessageReceiveTimeout (idle), immediately after a request is dequeued, and in the finally block after the request completes.
• New _validate_receive_timeout_vs_ping_timeout() classmethod raises ValueError at startup if the Redis transport's receive_timeout_in_seconds exceeds ping_timeout / 2, which would cause spurious kills of healthy workers blocked in a long BLPOP.
• main() accepts explicit _ping_timestamp and _ping_timeout keyword arguments so the values are available in spawn-mode child processes (Windows, macOS Python 3.13+) where sys.argv is reset.

Test coverage

• test_handle_next_request.py: _ping_parent no-op without shared memory, timestamp monotonicity, ping calls on idle/receive/complete/exception, receive-timeout validation (boundary conditions, Redis-default fallback, non-Redis skip), _ping_timeout kwarg priority over sys.argv.
• test_standalone.py: default fork count, CLI argument defaults, _ProcessMonitor watchdog (stale ping → SIGKILL, fresh ping → no kill, SIGTERM → SIGKILL escalation, graceful exit → no SIGKILL, respawn after ping-kill, stale-timestamp-from-dead-child doesn't kill replacement, startup timeout, startup window protection, ping timeout takes over after first ping).