Remove or improve the "error threshold" #83
Description
The threshold was introduced to accomodate for "random errors" (e.g. browserstack queue full, or random puppeteer timeouts), so as not to get noisy tests.
However it's not a great solution. In cases like https://circleci.com/workflow-run/fa1e2bc6-450b-4c3c-b648-40f51f0580b1, there's only one smoke tests, and it "errored" for a genuine problem - however the tests passed because it was within the threshold.
A better solution would catch and handle specific errors (like the browserstack queue one) and handle them specifically. For example, we could retry tests if it's a queue timeout. Or have specific types of errors that we do allow to fail. And fail for all other errors.