This diff introduces a call on the DagsterInstance to allow it to optimize for being the central copy used in the long running dagit process. This allows us to control two behaviors
- connection re-use / pooling
- statement timeouts
that we only really *know* what we want if its the copy used by dagit. We don't want long running connections on random instances of DagsterInstance, and we don't want an aggressive statement_timeout breaking things like instance migrations when used in the CLI.
Since its difficult to thread this option in at construction time due to the existing DagsterInstance / configurable class / instance ref machinery, adding this opt-in function was my best idea. Open to other suggestions.