P2: services._generation_kwargs now reads runtime.settings.horizon_days
instead of re-implementing the bool/non-int/<1 fallback against
runtime.options; drop the now-unused CONF_HORIZON_DAYS/DEFAULT_HORIZON_DAYS
imports. Remove NestQuestRuntimeData.options (kept only for that read; the
admin-id path reads entry.options directly) and the matching test assertion.
P3: _run_horizon_materialization's settings param is now REQUIRED (drop the
'| None = None' default + the NestQuestSettings() substitution branch);
update the test_timezones callers to pass NestQuestSettings() explicitly.
P3: __post_init__ normalises notify_target via _resolve_optional_str (strip)
so a hand-built NestQuestSettings(notify_target=' x ') behaves like the
options flow; keep the docstring accurate.
P3: delete the custom_components/nestquest/settings.py shim (settings is a
new module with no legacy import path) and import NestQuestSettings from
.core.settings directly in __init__.py, in alphabetical position.
P3: harden the AST scan for importlib.import_module('homeassistant...')
(string-based import) and call keyword names like hass= (ast.keyword);
add negative-control cases for each; explain in the failure message why
.config/.options/.data are banned (HA config-entry attributes with no
legitimate core use).
P3: drive from_options and from_options_resilient from one _FIELDS table of
(field, key, default, validator) to avoid three-way drift.
uv run pytest -q: 1573 passed.
overseer
d75ddc19bd