refactor: Rework internals

[Ehesp]

Purpose

As discussed on Discord, this is a pretty large change to allow full configurability of Starbase and overhauls some of the internals.

Key changes:

• The internal DO is now passed around via DataSource.rpc as a single RPC Session, this means for the duration of a single request, we're only charged for a single DO request, whereas currently it's multiple.
• Internals now have no reference to the environment or request, this is now provided as configuration to the handler (now called StarbaseDB.
• The handler now accepts a features object (to be documented) where you can disable/enable certain features.
• The handler now accepts a reworked DataSource instance, with:
  • A rpc property. This is the result of calling the DO init() method to return an RPC session.
  • A source, either external or internal. For worker deployed users, this is still set via the X-Starbase-Source header.
  • A cache boolean property - works is set to true and source is external. For worker deployed users, this is still set via the X-Starbase-Cache header is true.
  • A context property, which is what the RLS rules use. For worker deployed users, decodes the JWT and puts in the payload data. But for custom implementation, they can get the context from wherever they want.
  • An external property, which accepts a ExternalDatabaseSource type, which has all of our providers nicely typed.
• The expireCache logic has been moved into the internal handler.
• The handler now accepts a role, of either admin or client. Admin role does not apply allow list or RLS rules. For the deployed worker, this is set by observing the env api keys which are set.

Some additional notes / details:

• I have removed the MongoDB connection for now, since it doesn't support allow list or RLS due to the syntax differences. Let me know thoughts on this.
• I'm not sure if we can simply do this: new_sqlite_classes = ["StarbaseDBDurableObject"]? Do we have to tag the migration somehow?
• I noticed the current implementation checks the source on the URL when connecting via websocket... Shouldn't this be per query on the message level?

I've roughly tested things work.. but with no testing I'm conscious this is a big PR with lots of room to go wrong 👀

[Brayden]

I've added a few follow-up commits after testing all of the features. Overall this PR is a quality of life improvements introducing these high-level items, but many more (see original description) included as well.

• Type safety
• Ability to toggle on/off features
• Shared RPC reference for reduced costs