Old boundary clients do not send session_id. Instead of returning a
hard error that silently drops all logging and usage tracking, fall
back to log-only mode when session_id is absent or unparseable. DB
persistence is skipped but structured logging and the usage tracker
still run.
Update the agentapi unit tests to expect success (not error) for
missing and invalid session_id cases, and convert the agent e2e test
to a table test covering three client variants: old client (no
session_id), new client with correlation disabled (empty session_id),
and new client with a valid session_id.
Sas Swart
5d309eced2