Fixed issue with multiple asyncio loops on Windows.

The biggest issue was the construction of an asyncio.Semaphore() while the
default loop is alive, and then creating a new loop on win32.

I've also taken the opportunity to explicitly pass our loop to some calls,
rather than expecting them to use the correct one automagically, and added
some more explicit timeout handling to the semaphore usage.

blender_cloud/async_loop.py

@@ -44,6 +44,7 @@ def setup_asyncio_executor():
     executor = concurrent.futures.ThreadPoolExecutor()
 
     if sys.platform == 'win32':
+        asyncio.get_event_loop().close()
         # On Windows, the default event loop is SelectorEventLoop, which does
         # not support subprocesses. ProactorEventLoop should be used instead.
         # Source: https://docs.python.org/3/library/asyncio-subprocess.html
@@ -54,6 +55,10 @@ def setup_asyncio_executor():
     loop.set_default_executor(executor)
     # loop.set_debug(True)
 
+    from . import pillar
+    # No more than this many Pillar calls should be made simultaneously
+    pillar.pillar_semaphore = asyncio.Semaphore(3, loop=loop)
+
 
 def kick_async_loop(*args) -> bool:
     """Performs a single iteration of the asyncio event loop.