Tis PR ensures that the benchmark runs on the master branch. This makes benchmarks
executed in PRs compare directly against HEAD.

Also, the 150% margin for triggering an alert has a compounding degradation problem:

1. If a benchmark detects 149%, it doesn't notify and the test passes normally.
2. If the next benchmark also has a 149% loss, that loss is on top of the "new 100%", resulting in ~222% (1.49 × 1.49) accumulated performance loss between two distinct PRs, for example.

The margin was reduced to 110% (for now), limiting this silent drift.