V135: Structured staff skills.

Replaces the free-text phoenix_kit_staff_people.skills column with a first-class, translatable Skill entity assigned to people many-to-many, each assignment carrying zero or more of the skill's own proficiency levels. Creates:

• phoenix_kit_staff_skills — translatable skill (name + description + translations JSONB), globally unique by lower(name). Carries its own per-skill, translatable proficiency levels in a levels JSONB array (each {"id", "name", "translations"}) plus an allow_multiple_levels boolean that decides whether an assignment may hold one level or several.
• phoenix_kit_staff_person_skills — person ↔ skill join whose proficiency_levels JSONB array holds the selected level ids into the parent skill's levels ([] = no level / "not set")

Also adds a partial index on phoenix_kit_staff_people(date_of_birth) (active + non-null DOB only) so Staff.upcoming_birthdays/1 scans a small index rather than the full people table.

Data migration

The free-text skills column (comma-separated) is split, trimmed, case-insensitively de-duplicated into Skill rows, and each person is linked to the skills parsed from their string (proficiency NULL). Then the column is dropped. The parse/insert runs inside a column-existence guard so a partial re-run (column already dropped) is a safe no-op.

Lossy by design (documented): the column holds only the primary-language skill string. Per-locale skill overrides — translations[locale]["skills"] on the people table, a separate JSONB column — do not map cleanly to structured skills and are dropped (the orphaned "skills" keys are stripped from each person's translations). Structured skills carry their own translations going forward.