LIFX: improve light availability (#8451)

The default aiolifx timers are tuned for a network with few lost packets.
This means that lights can become "unavailable" from just a two second
dropout. An unavailable light is completely useless for HA until it is
rediscovered so this is an undesirable state to be in.

These tweaks make aiolifx try harder to get its messages through to the
bulbs, at the cost of some latency in detecting lights that actually are
unavailable.

homeassistant/components/light/lifx.py

@@ -37,6 +37,11 @@ REQUIREMENTS = ['aiolifx==0.5.2', 'aiolifx_effects==0.1.0']
 
 UDP_BROADCAST_PORT = 56700
 
+DISCOVERY_INTERVAL = 60
+MESSAGE_TIMEOUT = 1.0
+MESSAGE_RETRIES = 8
+UNAVAILABLE_GRACE = 90
+
 CONF_SERVER = 'server'
 
 PLATFORM_SCHEMA = PLATFORM_SCHEMA.extend({
@@ -117,7 +122,10 @@ def async_setup_platform(hass, config, async_add_devices, discovery_info=None):
     server_addr = config.get(CONF_SERVER)
 
     lifx_manager = LIFXManager(hass, async_add_devices)
-    lifx_discovery = aiolifx.LifxDiscovery(hass.loop, lifx_manager)
+    lifx_discovery = aiolifx.LifxDiscovery(
+        hass.loop,
+        lifx_manager,
+        discovery_interval=DISCOVERY_INTERVAL)
 
     coro = hass.loop.create_datagram_endpoint(
         lambda: lifx_discovery, local_addr=(server_addr, UDP_BROADCAST_PORT))
@@ -287,6 +295,9 @@ class LIFXManager(object):
             self.hass.async_add_job(entity.async_update_ha_state())
         else:
             _LOGGER.debug("%s register NEW", device.ip_addr)
+            device.timeout = MESSAGE_TIMEOUT
+            device.retry_count = MESSAGE_RETRIES
+            device.unregister_timeout = UNAVAILABLE_GRACE
             device.get_version(self.got_version)
 
     @callback