CASESessionManager now can mix up node ids from different fabrics

[bzbarsky-apple]

Problem

The fabric association was removed from CASESessionManager in #12766, but:

1. There are now callsites where we have a PeerId, we extract its NodeId and pass it in and reconstitute a PeerId.
2. There are still places where CASESessionManager uses bare node ids (e.g. FindExistingSession!) and these can now collide across fabrics, allowing something that should be talking via one fabric to talk via another fabric, possibly to a compeletely different destination. I would be shocked if there were not a security bug in here.

Proposed Solution

Change all CASESessionManager APIs to take const PeerId & instead of a NodeId. Drop the FabricInfo arguments. Do the fabric->GetPeerIdForNode(nodeId) in the consumers that don't already have a PeerId.

@msandstedt

[msandstedt]

Thanks @bzbarsky-apple for identifying this.