The timeout in the bisection is based on the run time of a good revision times two - matching the daca time regression report. But if the package runs very long the timeout might be too big.

Example (short-running):
If an analysis finishes very fast it might not exceed the timeout but might still show a sizable regression. For that case a run time factor threshold was introduced so be able catch differences which cannot be properly captured in an integer value.

Example (long-running):
If an analysis takes 30 seconds it requires 60 seconds to hit the timeout. But an increase in 5 or 10 seconds is already sizable. For that case a run time difference threshold has been introduced.

These changes also help with cases where the increase is incremental over various commits and not absolute caused by a single one.