[Doc] A some kind of "overview"/"illustrated" documentation to explain the architecture of GraphScope

**Is your feature request related to a problem? Please describe.**

I just realized that our existing documentation is focused on "Usage", i.e., "How", rather than "What". A explain/illustrated section in documentation (with some flow graphs, such a doc is not for general users, but for who want to understand how graphscope are organized and how it actually works) would be helpful for users or who are interested in GraphScope to gain a sense of

+ How "one-stop" are achieved, i.e., how the there compute engine co-works
+ The relations between components (client, coordinator, frontend, executor, engine, etc.) in graphscope, and how they talk to each other
+ How graphs are stored in the system
+ What the parameter `engine_cpu`, `engine_mem`, `vineyard_shared_memory`, etc. means, and why we need such resources
+ What will happen when users type `g = graphscope.load_ldbc_graph()`
+ Does graphscope achieve fault-tolerence, and what will happen when some nodes shutdown
+ Performance expectation and estimation
+ I have a graph (in csv or whatever), how should I specify resources when launching cluster (**_without talking to the internal developers_**)

And, it would be helpful for potential developers/contributors to gain a sense of where the chances existing that they can do something to contribute to graphscope.