--junitxml report includes time spent on migrations for first test

When using pytest's --junitxml flag to produce a test report, the timing information includes the migrations for the first test run which requires migrations. In our environment, we often randomize the order of test runs and split them based on past timing information. This doesn't work very well, because we can't measure the time taken to run the test when it includes the time spent on migrations – a time penalty we'll have to pay if any test uses the database fixture.

I'm not sure if there is an easy way to solve this -- but if there is, I'd happily take a stab at a PR.

Start by tracing pytest-django's handling of the Django database fixture and migrations alongside pytest's --junitxml timing. Determine where migration time enters the first test's reported duration, then verify that the report measures the test without that setup penalty while preserving correct timing for later tests.