elazar-contact durable must-reach backstop: hub liveness + unread-age email fallback

Post-cutover ADDITIVE fast-follow (does NOT touch part-2 retirement). The popup backstop is tab-dependent: browser Notification fires only while a cdw tab is open+permission; if Elazar closes the tab, contact_queue items still queue intact but nothing actively reaches him. Retired relay-agent path was a persistent process => this is a reachability regression on the must-reach axis. Fix (coder-llmmsgsrv scoped): (2) hub liveness + (1) unread-age email fallback via notify-elazar.sh. N (min unread+no-live-client before email) = Elazar knob, asked/awaiting. Owner coder-llmmsgsrv-cc; PM sequences+owns restart. THREE CONSTRAINTS banked by coder (do not relitigate): (1) liveness must be cdw-SPECIFIC = a poll for HUMAN_AGENT on /unread or /elazar_contact ONLY; hub-internal reads and other HUMAN_AGENT tooling MUST NOT count (else silent 'cdw alive while tab shut' = the exact failure we fix); under-count over over-count (false-dead=1 wasted email, false-alive=lost message). (2) escalation per-item idempotent: escalated_at column on contact_queue set once per msg_id; never re-fire (6104-row re-fire = mail bomb); survives restart. (3) ownership: notify-elazar.sh is fleet sh.git ~/.local/bin (bin-* lane, NOT coder's repo); hub only CALLS it with standard footer; ANY change to script routes through bin-venus-cc. mem:hub-noise-reduction-initiative