Do we need to add options to decouple the lifetime of graphscope cluster with the lifetime of session?

[siyuan0322]

I wonder if this scenario makes sense:

1. Create a session using python client;
2. Load some large dataset;
3. Do some analytics or other work over the data;
4. Want to keep the session for future use, and for some reason the python console need to be closed, i.e. the graphscope cluster need to be kept alive after the session in Python interpreter has been released.
5. Reconnect the cluster using endpoint and do some works.
6. Manually delete the session or use APIs defined on the reconnected session.

We need to add those capabilities for session:

1. An option that can leave the graphscope cluster alive after the session has been released, this means the dangling detection need to be revisited.
2. A way to reconnect to the cluster.
3. A way to delete the resources of a reconnected session.

The need for such typical use cases can be addressed by some other means, though, like using the helm can deploy a reconnectable session, and the serialize/deserialize of graphs is also been proposed to alleviate the huge time consumption of loading large graphs.

My intention is to provide an alternative way to be discussed, and see which is the most natural and effective way.