LargeDependencyGraphPerformanceTest.groovy

The test project doesn't work with the new "strict" constraint

semantics.

#10348 has an influence on some of these performance scenarios.

(correcting 90e852d and 85933bb)

Allow using Gradle profiler in cross version tests

Looking at the history, some regressions exist before branching. It's pointless to rebaseline master

but not rebaseline release for these scenarios.

To make clear that they are using the Gradle build internal

infrastructure.

There seems to be tiny gradual regression in performance, which causes unbearable

flakiness. Rebaselining all scenarios to catch future regressions.

This adds locking as an axis for performing resolution on a large

dependency graph.