Flag ruled-out approaches as possibly stale when their named file changed

[Coder8124]

Summary

• A ruled-out approach recorded against a commit is flagged as possibly stale at resume and before_you_try when the files its text names have changed since
• Moves the continuity eval suite out into the sibling logos-bench repo, leaving only the LongMemEval recall/QA self-check and bench pipeline here

Test plan

• go build ./... && go test ./... && go vet ./... && gofmt -l .

[Coder8124]

looks good for most part, had to divest the bench into another repo because it was conflicting with the product itself.