[BUG] @datagen_overrides seed is sticky when it shouldn't be

[abellina]

This bug has been there since the introduction of @datagen_overrides. Because of the global state in conftest.py the seed has to be reset each time a test instance is configured to an initial value, otherwise we keep the old datagen seed. The bug is that currently this seed isn't being reset to the initial value, so if you had a test with @datagen_overrides that preceded your test that didn't override, you'd still get this overwritten value.

To make matters worse, DATAGEN_SEED_OVERRIDE=X would still show. This is because this has to be rendered ahead of time in order to allow for tests that are divided into multiple python processes (xdist). So it followed a different path, showing the incorrect seed.

I found this while working on #10089 by adding a print statement in conftest.py:get_datagen_seed. This function is called from datagen.py to get the seed. I found this was returning the wrong value.