Add a max_attempts to exp_delay

This is minor, but the current exp_delay (used for exponential backoff in core loop) has a timeout.  We should have an option, at least, for a max attempts instead/in addition.  The core loop times out itself, so the exponential delay need not enforce that.  But it is nice to have a backup plan which at least keeps it from spinning forever.   Currently, it times out at 2x the expiration_time for the txn.  That's fine, for now, but fairly cheesy.

Also - when this does happen, we need to deal with the exception in a way that we are sure the logging will make clear what happened.