Required input: Tag expression to select a single testcase to investigate.
Required input: commit hash of good/old VPP commit.
Optional input: commit hash of new/bad VPP commit (if not HEAD of job's branch).
Optional input: Parameter overrides for the test.

Operation: Build bad VPP, run a single BMRR testcase (without CRC checks). Build good VPP, run MRR tests. Start "git bisect". Repeat:
Build VPP on commit chosen by git bisect, run BMRR test. On failure, "git bisect skip". Use jumpavg to add the result either to the group of good or bad results. Tell git bisect good/bad verdict. Continue until git bisect finishes.
Finally, "git bisect reset" to return the repo to initial state.

Open questions:
Store intermediate results in Python objects or files managed by bash script?
Can "git bisect run" be used? If not, how to detect bisecting has finished?