async_added_to_hass listed the switchable boxes inline, one fetch per box
popup, so on a slower connection the platform setup ran past
SLOW_SETUP_MAX_WAIT and the whole config entry was cancelled into
setup_error — every entity unavailable despite valid credentials. Move the
listing to a background task so setup never blocks on it; the options fill in
once it returns.
Also bound every portal request to a 30s timeout so one hung request can no
longer drag a refresh, or a first setup, past Home Assistant's own limits and
fail it outright.
Regression covered in tests/test_setup_hygiene.py.
Three fixes the round trips found:
The produce-box select reported the SUBSCRIPTION rather than the delivery. A
one-off switch changes the delivery while the standing order keeps naming the
old box, so during the exact week someone had changed it the entity showed the
wrong box. It now reads the order.
parse_vacation_holds looked for ISO dates. The page writes 'Tuesday, Dec 1 -
Monday, Dec 7', so it reported no holds on an account that had one — which is
indistinguishable from having none. Its test asserted the same wrong format,
so the test passed while the parser was blind.
Adds hold removal (POST /s/submit/pause-range-remove), whose popup only exists
while a hold does.
The to-do list exists so Home Assistant's own conversation agent can manage the
order through HassListAddItem rather than through anything bespoke. Skip is a
switch because skipped/not-skipped is state worth reading back; donate is a
button because it cannot be undone. Un-skip raises rather than guessing at an
endpoint that has never been observed.
Fixes subscription parsing, which matched the heading row and so reported zero
on an account that has one.