- open transport cooldown only after timeout retries are exhausted so
bounded retries keep their short backoff instead of waiting out a
full 30s cooldown in the rate guard
- shield streaming billing finalization and upstream connection close
from client-disconnect cancellation
- stop proof selection at the minimal covering set when over budget so
the re-quote shortfall is not inflated by extra input fees
- reuse is_mint_transport_error instead of a duplicate helper in lnurl
- ppqai: normalize circuit origin default ports, drop dead try/except
in fetch_models