Unsubscribe dispatcher and update listeners on remove/unload

[andreaslindroos]

What's broken

Every config entry reload leaks the old sensors. They are removed from HA,
but they keep recalculating on every quarter-hour tick. With enough leaked
sensors this blocks the event loop and freezes all of HA for minutes at a
time: automations, UI and integrations all stop.

It fails silently. Nothing is logged: no error, no warning, and HA's
blocking-call detection doesn't catch CPU-bound work. All you see is HA
going unresponsive every 15 minutes, with MQTT clients timing out as a side
effect, and nothing pointing at nordpool. This also affects automations and many
probably have automations running at exact hour or half past or similar.

Why

• async_added_to_hass calls async_dispatcher_connect for NEW_DAY,
  NEW_PRICE and NEW_HOUR and discards the unsubscribe callbacks, so
  removed sensors stay subscribed.
• The update listener is never removed, and async_reload_entry calls
  unload/setup directly. Listeners stack up and later updates trigger
  several reloads, each leaking more sensors.

Example of this happening

HA froze for ~2 min at every :00/:15/:30/:45, with nothing in the logs.
Stack dumps showed the event loop in handle_new_hr → _update →
_calc_price for the whole freeze. One render takes ~0.3 ms, but there
were 133 live sensor objects for 3 real sensors.

Fix

• Wrap the dispatcher connections in self.async_on_remove(...).
• Register the update listener with entry.async_on_unload(...).
• Reload via hass.config_entries.async_reload(entry.entry_id).

Tested: after a restart with this patch, the nordpool_update_hour
subscriber count stays at 3 across reloads, and the freezes are gone.