Fix bug in Trio's _read_exactly()

This method calls Trio's `receive_some(n)`, which I found out can actually
return fewer than `n` bytes. This is only noticeable on large messages
(query results of around 80KB triggered the issue). The solution is to call
`receive_some(...)` in a loop until `n` bytes has been received.

Author: mehaase

rethinkdb/trio_net/net_trio.py

@@ -237,8 +237,11 @@ async def _read_until(self, delimiter):
         return bytes(buffer)
 
     async def _read_exactly(self, num):
+        data = b''
         try:
-            return await self._stream.receive_some(num)
+            while len(data) < num:
+                data += await self._stream.receive_some(num - len(data))
+            return data
         except (trio.BrokenResourceError, trio.ClosedResourceError):
             self._closed = True