maxDelay bounds the wait and sets the bar a link must clear, and only the
first was documented, so tuning it up quietly lengthened every recovery.
The unreadable PDU the tests write is one named constant now.
A bind that returns is not proof the link works: an unreadable stream is
only found afterwards, so resetting the delay there handed a link that
died on arrival a fresh minDelay every cycle — one connect and bind per
second forever, which is how an account gets blocked for bind flooding.
The loop resets only once a link has outlasted maxDelay.
Also covers the unreadable stream that has no loop to retry it.
A framing or codec error tears the link down rather than the session, so
the loop retries it on a fresh socket with a fresh framer — which is what
a desynced stream needs. Removing that reset failed nothing before; the
reconnect test now leaves half a PDU on the dying link, and does.
`disconnected` counts failed links rather than outages, which the README
now says, and the transport wires its socket as it is built.
Ending a session while the link was down emitted nothing: the retried
drop had already torn it down, so close() found nothing left to do. The
terminal event is now guarded by its own flag rather than by `closed`.
The socket, the framer and the parse move to PduTransport, which is what
takes session.ts back under its line cap; `session.sock` reads through a
getter. The backoff also refuses an inverted pair and an unknown key.
`close` fires only where nothing will bring the session back; a drop the
reconnect loop will retry is `disconnected`, pairing with `reconnected`.
`minDelay: 0` never doubles, so the backoff never started — both delays
are range-checked now, and only `false` spells reconnect off.
Surviving a drop is most of what the session layer is for, and behind an
opt-in an application that never read the options table got none of it.
`reconnect` still takes `{ minDelay, maxDelay }` to retune the backoff;
`false` is the one spelling for off.
lilleman