feat(presence): register a vacation/absence via calendar.addVacation

[nickknissen]

What

aula vacations reads vacation registrations, but there is no way to create one. The portal has this write, and it is the correct tool for a planned absence, which report-sick deliberately does not cover (Aula has no future-dated sick report).

Grounding

Portal bundle, presence Vuex store (https://www.aula.dk/static/js/0.*.js, webpack module 152), sitting directly beside the UPDATE_SICK action:

ADD_ABSENCE: ({commit}, payload) =>
  post("?method=calendar.addVacation", payload)
    .then(r => { commit(MUTATE_SUCCESS_TEXT, "SUCCESS_TOAST_REGISTER_ABSENCE"); return r })
    .catch(() => commit(MUTATE_ERROR_TEXT, "API_ERROR_ADD_VACATION"))

Note the method lives under calendar.*, not presence.*, even though the presence UI is what calls it.

The mobile app decompilation (com.netcompany.aulanativeprivate, published in eisbaw/aulalibre) lists the same family: AddVacation, GetChildrenVacation, GetVacationAnnouncementsByChildren, GetVacationRegistrationOverview, GetVacationRegistrationsByChildren.

Work needed

The request payload is not pinned down yet. Before implementing, read it out of the bundle the same way the sick write was grounded in #63:

1. Find the component dispatching ADD_ABSENCE and record the exact object it passes (child ids, start/end dates, a note field, and how the vacation "announcement" case differs from a plain registration).
2. Cross-check against DTOs/Models.ComeGo vacation entries in the app decompilation's data_models.md.
3. Check the permission gate: PresenceModule.VACATION already exists in models/presence.py, so PresenceConfiguration.can_edit(PresenceModule.VACATION) should gate the command exactly as report_sick gates report-sick.

Then: AulaApiClient.add_vacation(...), a CLI command (aula register-absence or a --add mode on aula vacations), the confirmation and --yes pattern used by the other writes, and tests covering the payload plus the permission skip.

Context

Found while grounding the sick report in #63. Deliberately left out of that PR to keep one write per change.