Currently db_repl closes an execution session via on.exit().

It's possible to adjust the behaviour to have sessions be re-attachable if the computes still present.

Points to consider/discuss:

• Are sessions aliased with a name to make the UX easier?
• Are sessions managed on disk in a cache or is it on the user to remember the Id?
• ContextId's are long and unfriendly - maybe theres a way to turn that into a memorable name?

Having resumable contexts would play well with potential integration to {mcptools} and my idea to support this within {brickster} #111 and could allow any command to be run on Databricks by tools like claude code etc.