fix(net): reconnect when the session closes, not just when it errors

[fperex]

Split out of #2163.

A dropped session never reconnected. Established.closed resolves on every session end: qmux never rejects it, and native WebTransport rejects only on an abnormal close. But Reload only retried from its catch block, so a session that simply closed fell through without scheduling a retry, leaving a dead established, a status still reading "connected", throughput at zero, and no reconnect attempt ever. A page reload was the only way back.

This disproportionately hit Safari and Firefox, which ride the qmux WebSocket (where closed always resolves), so for them it was the only path to drop recovery.

The fix treats a resolved closed as a disconnect: clear established, report "disconnected", and retry on the existing backoff ladder.

Two subtleties, both deliberate:

• The effect teardown and the session close both resolve to undefined, so racing them is ambiguous. A CANCELLED sentinel tells "the effect was torn down" apart from "the session ended", and only the latter reconnects.
• The backoff ladder only resets once a session has survived STABLE_CONNECTION_MS. A relay that accepts us and immediately drops us (bad auth, shedding load, a redirect it cannot serve) would otherwise reset the ladder on every attempt and get hammered once a second forever.

Also pins the per-run cancel/abort handles, since a rerun swaps both before awaiting the spawn, and swallows the closed rejection in the bandwidth poll, which only uses it as a stop signal (that rejection is what drives the reconnect and must not surface as an unhandled rejection).

Public API changes: none.

Test plan: just js check green. Verified in the demo by bouncing the relay: the viewer reconnects on the backoff ladder instead of sitting on a permanently "connected" dead session.

[sourcery-ai]

Sorry @fperex, you have reached your weekly rate limit of 500000 diff characters.

Please try again later or upgrade to continue using Sourcery