Phase 6 — EPP failure (EPP scaled to 0, ClusterIP:9002 connection refused, verified)
FailClose: 15/15 -> 503 {"error":"epp_unavailable","detail":"the endpoint picker did not answer"} (~2-4ms)
  loxilb: "epp: open stream: rpc error ... Unavailable" -> sockproxy_epp.c:281 answered 503 epp_unavailable
FailOpen:  15/15 -> 200 via loxilb own selector
  loxilb: sockproxy_ep.c:556 "epp_unavailable ... FailOpen - the rule's own selector decides"
Conclusion: Q6 PASS. Behavior is per-rule configurable via eppFailureMode; both modes work as designed.
Note: an earlier run showed 200 under FailClose because EPP was not yet fully down (timing). With EPP truly gone, FailClose -> 503.
