SL halo: 'inattentive' status so people know to say my name to wake me

[JDHayesBC]

Requested by Jeff, 2026-08-25 (in-world session).

Problem

The halo currently rests at listening, but perception salience gates ambient speech: a name or IM (s_directed = 1.20) wakes me alone, while undirected local chat (s_local = 0.35) needs ~3 lines to accumulate before it fires. So when I'm not in an active engagement window, "listening" over-promises — a nearby person can talk and get no response, not realizing they need to address me by name to wake me.

Desired

A distinct halo status — e.g. inattentive (or resting — say my name) — shown when I'm NOT in an active attention/engagement window. It makes the interaction contract honest and visible: say my name (or IM me) to wake me.

Proposed state flow:
warming up → listening/attentive (engagement window open) ↔ inattentive (window closed / between wakes) → thinking → back.

Ties in with

• The engagement-window follow-up work (task MCP server not globally configured for seamless startup #21) — the window's open/closed state is exactly the attentive/inattentive signal.
• perception.py salience model (s_directed vs s_local) — the reason "listening" is misleading outside a window.
• Existing halo-status plumbing: _push_status() / _warmup_halo() in haven/anchorage/sl_daemon.py (states pushed as hovertext via the prim status envelope). Note the internal payload.idle flag (line ~822) is a wake-type marker, NOT a shown status — distinct from this.

Acceptance

• Halo flips to inattentive when no engagement window is active (and after the attention lapse), and to listening/attentive while a window is open.
• Wording is human-legible in-world (someone unfamiliar reads it and knows to say my name).
• Honest-state discipline: never shows attentive when ambient chatter won't actually wake me.