async Client with sync LocalCluster

When a Client is passed a LocalCluster and the asynchronous=True flag, it silently ignores the explicit flag and gets it from the cluster instead.

```python
>>> cluster = await distributed.LocalCluster(asynchronous=True)
>>> client = await distributed.Client(cluster, asynchronous=True)
>>> client.asynchronous
True
>>> cluster = distributed.LocalCluster()
>>> client = await distributed.Client(cluster, asynchronous=True)
>>> client.asynchronous
False
>>> cluster = distributed.LocalCluster()
>>> client = await distributed.Client(cluster.scheduler.address, asynchronous=True)
>>> client.asynchronous
True
```
The same happens for an eventual explicit loop parameter.
Both issues are caused by these lines, which cause Client.loop to be forced to match Client.cluster.loop:
https://github.com/dask/distributed/blob/9f91653398aee681ba0010b1053f3bd1b415894b/distributed/client.py#L647-L653

I see in the code how a few things will break if the client and client.cluster end up having two different event loops. Not sure what the correct solution would be?