Remote comms: Handle reconnection to previously-dead peers with incarnation ID detection

[sirtimid]

Problem

When a connection is marked as permanently dead (no longer retrying), a user action that initiates a new connection to the same peer should:

1. Clear the permanent failure state and allow the connection
2. Use incarnation ID logic to detect if it's the same peer instance or a restarted peer
3. React appropriately based on whether the incarnation changed

Current Gaps

1. sendRemoteMessage() doesn't clear permanent failure - if a user sends to a permanently-failed peer without calling reconnectPeer() first, the message will fail
2. Handshake not integrated into connection flow - handshake module exists (handshake.ts) but channels are registered without exchanging incarnation IDs
3. No incarnation change callback - when incarnation changes (peer restarted), RemoteHandle state needs to be reset

Proposed Solution

Part 1: Clear Permanent Failure on User-Initiated Send

In sendRemoteMessage(), check and clear permanent failure before dialing.

Part 2: Integrate Handshake into Connection Flow

• Add localIncarnationId parameter to initTransport()
• Perform outbound/inbound handshake after channel establishment
• Pass handshake deps to reconnection lifecycle

Part 3: Add Incarnation Change Callback

• Add OnIncarnationChange callback type
• In RemoteHandle, add handlePeerRestart() method to:
  • Reject all pending messages
  • Reject pending URL redemptions
  • Reset sequence numbers
  • Clear persisted sequence state

Part 4: Same Incarnation (Resurrection)

No changes needed - normal operation continues with existing sequence numbers.

Files to Modify

• packages/ocap-kernel/src/remotes/platform/transport.ts
• packages/ocap-kernel/src/remotes/platform/reconnection-lifecycle.ts
• packages/ocap-kernel/src/remotes/kernel/RemoteHandle.ts
• packages/ocap-kernel/src/remotes/kernel/RemoteManager.ts
• packages/ocap-kernel/src/remotes/types.ts
• packages/ocap-kernel/src/store/index.ts (may need clearRemoteSeqState())

Behavior Summary

Scenario	Behavior
Send to permanently-failed peer	Clear failure, dial, handshake, proceed
Same incarnation ID	Normal operation, pending messages may still be ACKed
Different incarnation ID	Reset RemoteHandle state, reject pending, start fresh