sip: retry IMS registration instead of silently going dark

[SavedByLight]

Previously any REGISTER failure (bad challenge, non-200 response, or a thrown exception in registerCallback via require(statusCode == 200)) left the client permanently deregistered with no automatic recovery path other than the periodic keep-alive alarm, which only fired every 50 minutes. Combined with onAvailable()'s one-shot network.isInitialized guard (which could never retrigger a connect() once set, even after the network was lost and regained), a single transient failure could mean missed VoLTE calls/SMS for up to 50 minutes, or indefinitely until a manual VoLTE/airplane-mode toggle.

SipHandler.kt:

• Add sessionActive flag, scheduleReconnect() with exponential backoff (5s -> 120s cap), and ensureRegistered().
• connect()'s two hard-failure paths (bad initial challenge, failed authenticated REGISTER) now call scheduleReconnect() instead of just returning.
• registerCallback() no longer throws/kills the thread on a non-200 response; it schedules a reconnect instead.
• A successful registration resets the backoff counter and marks the session active.
• getVolteNetwork()'s NetworkCallback now keys reconnection off sessionActive instead of the permanently-latched network.isInitialized check, so losing and regaining the network actually triggers a reconnect.

PhhImsService.kt:

• Shorten the periodic keep-alive alarm from 3,000,000ms (50 min) to 180,000ms (3 min); cheap now that ensureRegistered() only does real work when needed.

PhhImsBroadcastReceiver.kt:

• Periodic alarm handler now calls ensureRegistered() instead of a raw register() against a socket that may already be dead.