tests: when making a `memctx`, make sure to use a single repo instance The way self.repo is dynamic produces bad lock behavior because the `context.memctx` ends up with a different instance than self in `localrepo.commitctx`, which means the callbacks in the `memctx` get an unlocked repo instance. This causes lock warning failures. When it's not a code freeze for core, we should probably: * Make lock failures hard, not just warnings * Stop holding a repo reference in memctx, or otherwise check it's the same instance as `self` during `localrepo.commitctx` That's my best guess based on the (very hard to debug) test failures here.