perf(mobile): reconnect environments immediately on resume

[t3dotgg]

Returning to T3 Code Mobile after using another app could leave every suspended environment waiting on a socket probe and retry backoff, making multi-environment reconnects feel sequential.

This change gives mobile resume an explicit fast path:

• brief interruptions immediately probe existing sessions and refresh shell/thread subscriptions
• meaningful background suspensions replace stale session leases so environments reconnect in parallel
• foreground activation interrupts any in-flight probe or retry delay and resets the retry ladder
• a reconnect request received during an active probe is retained and honored after the probe finishes

The branch is rebased onto current main, including #4882. That PR improves mobile action/feed responsiveness but does not overlap the connection runtime. After comparing #4885, this PR adopts its useful retry-backoff reset while keeping the existing in-memory token catalog and relay synchronization model; migrating persisted tokens or widening the relay critical section would add failure surface without improving the normal foreground path.

Verification

• 58 focused client-runtime/mobile tests
• @t3tools/client-runtime typecheck
• @t3tools/mobile typecheck
• targeted client-runtime lint
• iOS Simulator: two connected environments plus five offline environments, backgrounded for 11 seconds, then resumed from the app icon; connected environments stayed available while only the offline environments retried

Watch the full verification clip (the middle background interval is accelerated).

Made with GPT-5.6 Sol in the Codex harness through T3 Code.

Note

Reconnect mobile environments immediately on resume after long background suspension

• Adds mobileApplicationActiveWakeup in app-state-wakeups.ts to classify resume events as either application-active-probe (short interruption, <10s) or application-active-reconnect (long suspension, ≥10s).
• On application-active-reconnect, the connection supervisor resets the retry backoff ladder and immediately replaces the session rather than probing the existing one.
• On application-active-probe, the supervisor runs a liveness probe with a 3s timeout (vs 15s normally); shell and thread state resubscribe on probe but not on reconnect.
• Bearer descriptor caching (10s TTL) and a 3s DPoP socket ticket timeout are added to the authorization service to reduce reconnect latency.
• Behavioral Change: application-active wakeup is split into two distinct kinds; any code relying on application-active alone will not receive mobile resume events.

^{Macroscope summarized e279787.}

---

Note

Medium Risk
Changes core connection supervisor behavior, mobile wakeup semantics, and remote auth caching/timeouts; broad test coverage but affects reconnect and token paths for all mobile environments.

Overview
Mobile resume no longer treats every foreground return like a single application-active probe. Short background gaps emit application-active-probe (3s health check, shell/thread resubscribe). ≥10s in background emits application-active-reconnect, which replaces the session immediately and resets the retry ladder instead of waiting on probes or backoff.

The connection supervisor honors application-active-reconnect during setup, while connected, and during an in-flight probe (probe is interrupted). Foreground activation also resets backoff when offline, blocked, or waiting to retry.

Remote authorization caches validated bearer descriptors for 10s (fewer descriptor fetches; fresh websocket tickets each connect). Cached DPoP endpoint use now uses a 3s websocket-ticket timeout and refreshes after the first transient socket failure (replacing a two-failure threshold).

Shell and thread sync only resubscribe on probe-style wakeups, not on reconnect (new session already drives sync).

^{Reviewed by Cursor Bugbot for commit e279787. Bugbot is set up for automated code reviews on this repo. Configure here.}