Resource leak when tube is removed and -all is used

I was reading the source and thought I would report a tiny bug such that no one runs into it:

[`watchNewTubes()` method](https://github.com/99designs/cmdstalk/blob/fbd704da919d283643e673bd02efcf0221224b92/broker/broker_dispatcher.go#L82-L95) only starts new worker dispatchers, but never closes down preexisting ones if a tube is removed. If you have a lot of tubes being created and destroyed, you will run out of memory eventually.

The issue points to the watchNewTubes() method in broker/broker_dispatcher.go. Start by understanding how tube removal is handled elsewhere in the codebase. Look for existing cleanup logic or worker shutdown patterns. Determine what state needs to be tracked to close dispatchers for removed tubes. Running the existing tests will help verify any changes.

Scritto dal modello di indicizzazione a partire dal testo della issue.