For both L1 nodes and L2 nodes, decouple the idea of a "chain" and a "node".

A chain should:

• have a way to access configuration
• have a way to access deployment info
• have a way to access nodes by name (and per type, e.g. getting an EL node, or getting a rollup-node)

An (EL) node should:

• provide high level interface, to interact with all common RPC client bindings, without leaking the client binding to the test itself.
  • direct RPC access in a test is bad, it's not DSL, it makes it harder to maintain and apply shared behaviors too.

Similarly, an interface for a rollup-node, around the rollup-client RPC bindings, is useful to have.
Start really minimal, interfaces and composition is important. We can expand the interfaces later.
E.g. the sync-status RPC.

Over time, preventing boilerplate in tests is important. E.g. an interface can provide "wait for unsafe head change", instead of making it poll the sync-status manually.

Also see ticket for expanding the EL op-service client bindings.