Stress test for ImmixAllocator should not change limits of the allocator

[k-sareen]

When I fixed stress test for ImmixAllocator I implemented require_new_block in a way that had a side-effect of attempting to recycle lines/blocks. Back then I thought this wouldn't cause issues but I believe I've recently triggered a bug in my port that is caused by this. Ideally the stress test should be side-effect free. The issue is that the current code isn't architected in a way that makes it easy to do. For example, the associated ImmixSpace may end up popping recyclable blocks and update the hole searching line field in the current implementation.

[qinsoon]

I don't quite get this. Can you explain what is the expected behavior, and what is the wrong behavior you observed?