Have SWE-smith challenge the model to generate *tests* #64
Description
Right now we only get the model to generate bug-fixes/feature-requests.
What if we could also get it to generate tests?
I think we should use an approach similar to SWT-bench https://swtbench.com/
where we have it generate a test and then test it by making sure the test fails before a gold-patch is applied and passes after the gold-patch is applied.
Testing is one of the most important aspects of SWE, so this enhancement could really drive up performance a lot.