python: release GIL while calling terminate to avoid deadlock

[crockeo]

Description: Alternative to #1379 for getting rid of the deadlocks. The deadlocks are caused by:

• Thread 1 calls Engine::terminate with GIL in hand, asks Thread 2, and then waits for it to die.
• Thread 2 (envoy main thread) begins teardown, which causes EngineCallbacks::~EngineCallbacks to run, which attempts to grab the GIL.
• Thread 1 is waiting for Thread 2 to die
• Thread 2 is waiting for Thread 1 to give up the GIL

Risk Level: Low

Testing: ...a little convoluted. These deadlocks occur at process teardown, so I've been installing the envoy-requests wheel and making a pair of files:

# named test.py
import envoy_engine
import envoy_requests
from envoy_requests.common.engine import Engine

if __name__ == "__main__":
    Engine.log_level = envoy_engine.LogLevel.Debug
    envoy_requests.get("https://api.lyft.com/ping")

and

#!/usr/bin/env bash

while true; do
    python test.py
    sleep 0.5
done

and then running several of these until a deadlock occurs. You can also reproduce them by running several test jobs at the same time (--runs_per_test), but I found the above method to result in a deadlock sooner.

Docs Changes: N/A
Release Notes: N/A

[crockeo]

unnecessary with #1379 merged