Add better logging to `UserDeleteService`

[seanh]

Demo:

Purging User(id=1) - marked user as deleted
Purging User(id=1) - added purge_user job: Job(id=3, name=<JobName.PURGE_USER: 'purge_user'>, enqueued_at=datetime.datetime(2024, 11, 20, 17, 54, 14, 532500), scheduled_at=datetime.datetime(2024, 11, 20, 17, 54, 14, 532500), expires_at=datetime.datetime(2024, 12, 20, 17, 54, 14, 532500), priority=0, tag='UserDeleteService.delete_user')
Purging User(id=1) - added record of user deletion: UserDeletion(id=1, userid='acct:devdata_user@localhost', requested_at=datetime.datetime(2024, 11, 20, 17, 54, 14, 532500), requested_by='acct:devdata_admin@localhost', tag='admin.users_delete', registered_date=datetime.datetime(2024, 11, 20, 17, 51, 45, 532818), num_annotations=2)
Purging User(id=1) - deleted 1 rows from authticket
Purging User(id=1) - deleted 2 rows from token: 1, 2
Purging User(id=1) - marked annotations as deleted: ZTAldqdoEe-rRre0o_POkw, ZyjywqdoEe-rRsdA2rr4SA
Purging User(id=1) - marked annotation_slims as deleted: 1, 2
Purging User(id=1) - enqueued jobs to delete annotations from Elasticsearch: ZTAldqdoEe-rRre0o_POkw, ZyjywqdoEe-rRsdA2rr4SA
Purging User(id=1) - deleted 1 rows from group: 5
Purging User(id=1) - deleted 1 rows from user: 1
Purging User(id=1) - completed job: Job(id=3, name='purge_user', enqueued_at=datetime.datetime(2024, 11, 20, 17, 54, 14, 532500), scheduled_at=datetime.datetime(2024, 11, 20, 17, 54, 14, 532500), expires_at=datetime.datetime(2024, 12, 20, 17, 54, 14, 532500), priority=0, tag='UserDeleteService.delete_user')

The aim is to add good enough logging to UserDeleteService such that if a user was accidentally deleted (or if a bug caused UserDeleteService to delete stuff that it shouldn't, etc) we'd have enough information in the logs to assess what happened and to recover the accidentally deleted data from a backup of the DB.

The logging should also incidentally make monitoring and debugging generally easier.

Notes:

• __repr__()'s on all model classes that're nice and short, include the primary key, and don't include any sensitive information. And then log model objects with %r or {obj!r} so that __repr__() is used not __str__().

• When rendering lists of things from the DB into log message strings it's good to sort the things. This makes the app's behaviour more predictable: always logging things in alphabetical order. And makes it easier to write non-flakey test assertions.

• It can be useful to have helper functions to log things in a consistent way. For example all the UserDeleteService log messages in this PR start with "Purging {user!r} - " and when DB rows have been deleted or updated there's always a comma-separated list of the IDs of the affected rows.