In WT-4174 I was able to reproduce a seg fault by adding an in-memory configuration to test_timestamp04.py. However committing that addition is not fully correct. It reproduced the bug early enough to work on, but adding that configuration to the test permanently requires significantly more work in the test to handle the potential of WT_CACHE_FULL errors at some point in the test, what that means for later parts of the test, etc.

So this is a separate ticket to improve testing in WT unit tests so that the code fix for the seg fault can unblock others.