[bug] whole-home fidelity guard rejects "turn on/off the lights" — on/off not benign, so a valid command fails to actuate

[ultrahighsuper]

Summary

entity_fidelity::whole_home_resolution_is_trustworthy treats the command verbs turn/switch/set/get as benign (they don't pin a request to a place), but NOT the state particles on/off/up/down that pair with them. So a valid whole-home command like "turn on the lights" fails the guard on the stray on token, the whole-home resolution is judged untrustworthy, and resolve_domain_target returns None — the lights never actuate.

Steps to reproduce

Resolve a bare-domain command through the runtime whole-home path (ha::provider::resolve_domain_target → whole_home_resolution_is_trustworthy):

• turn on the lights
• turn off all the lights
• turn on the kitchen lights

Expected

Trustworthy → the command resolves to the light group and actuates. turn is already benign; on/off should be too (they name no place).

Actual

whole_home_resolution_is_trustworthy(home, "turn on the lights") returns false because on is neither a domain word, a benign word, nor a valid place — so the guard rejects a correct resolution and the lights stay unchanged.

Notes

The BENIGN_QUERY_TOKENS list already whitelists the verbs (turn, switch, set, get, status), which shows it's meant to tolerate command-like text — it just misses the on/off/up/down particles. crates/genie-core/src/ha/entity_fidelity.rs. This is real-HA tool-dispatch/actuation correctness.

Hardware

Non-Jetson (x86_64 dev / cross-build host)

GenieClaw version / commit

2ad90f4 (main)