Adopt workspace_id / endpoint_addr / session_id split

[tanlethanh]

Summary

Host-side workspace identity needs to be split into three distinct fields:

• workspace_id: durable Zedra workspace ID stored in host storage
• endpoint_addr: transport endpoint address for that workspace
• session_id: the currently attached session for that workspace

This removes the current naming overlap where one concept is being used for multiple identifiers.

Scope

• Update host storage and workspace state naming
• Keep transport and connection identifiers separate
• Update docs and protocol notes to match the new convention

Notes

• session_token stays ephemeral and per-connection.
• This is a naming/model cleanup, not a behavior change.
• endpoint_addr is misnamed: despite the name, the string used everywhere (URL params, WorkspaceState.endpoint_addr, deeplinks, saved-state JSON) is the workspace identity, not a network address. It is the id-only encode_endpoint_addr(EndpointAddr::from(endpoint_id)) form (no relay/direct addresses), matched by equality for persistence/reconnect/deeplinks. Pitfalls this causes: encoding a full endpoint.addr() (with relay/direct addrs) where identity is expected yields a different string for the same node and breaks matching (was a real bug in zedra status); it is also distinct from the raw endpoint_id.to_string() (plain base32). The split should make the identity field's name reflect "id", separate from any true transport address. (Doc comments added at zedra_rpc::pairing::encode_endpoint_addr and the endpoint_addr fields in the interim.)