Record originating task ID as external_id

[snopoke]

Uses the new task external_id field (server taskbadger/taskbadger#413) to automatically record the originating framework's task ID, so a Task Badger task can be correlated with the Celery/Procrastinate job when digging through logs. Also exposes external_id on the public SDK (create_task/update_task/Task.create/Task.update), mirroring queue.

Worth knowing:

• Celery records the task ID for free at creation (headers["id"] / request.id). Procrastinate's job ID is DB-assigned and only known after defer() returns, so it's recorded via one follow-up PATCH at enqueue time (covers defer, defer_async, and periodic jobs).
• internal/ was updated by hand rather than committing the full invoke update-api regeneration: the pinned generator drifts unrelated code from dateutil.isoparse to datetime.fromisoformat, which can't parse a trailing Z on Python 3.10 (a supported version). The hand-edits match the generator's external_id output exactly.
• The Celery handler reuses a class-level taskbadger_kwargs dict across invocations; that dict is now copied before mutation so a per-publish external_id can't leak to the next publish (regression test included).

No version bump included — left for release time.