ci: Avoid second db entry in tests (#4749)

* Simplify experiment tests

* Remove breakpoint :(

* Reintroduce plateform skips

* remove pydantic-specific versions

* remove pydantic-specific versions

* Sleep a while

* Properly check score

* Fewer evaluators for consistency

* Remove needless empty dataset check

* Uncomment flaky skip

* Remove flaky assertion

---------

Co-authored-by: Alexander Song <axiomofjoy@gmail.com>