Add a way to set allow_cascade=True in _fixture_teardown()

[Howard Cox]

We use ​https://github.com/schlunsen/django-timescaledb as our database backend and ran into a problem with some of our tests for django migrations on tables using timescale compression.

I [wrote up our problem on django-timescaledb back in October (​https://github.com/schlunsen/django-timescaledb/issues/40) but it essentially boils down to:

• django.test.testcases.TransactionTestCase._fixture_teardown() calls django.db.backends.postgresql.operations.sql_flush() and truncates every table at the end of tests.
• In timescaledb, compressed hypertables create "chunks" as tables with a subset of the hypertable's data.
• Because we have foreign keys in some of our hypertables, the chunk tables also get that foreign-key rule.
• Those chunk tables are transparent to Django, and so are not in the list to be truncated.
• And so, the truncation fails, because the chunk tables still reference some of the data being truncated.

The easiest solution seems to be to allow _fixture_teardown() to opt-in to setting allow_cascade=True when truncating.

We're happy to submit a PR if this proposal gets accepted.

[Mariusz Felisiak]

As far as I'm aware, it's an issue in timescaledb, they could override ops.sql_flush() and add flushing "chunks".

[Howard Cox]

That's fair, it's been abandoned anyway so we're considering forking it, so can just change it.