collect the human-written commits that follow the failing commits and pass all tests

[martinezmatias]

To compared the generated patches with those created by the developers, it could be useful to collect the commits that follow the commits that produce the build failure on Travis. When Repairnator detects a passing build for an app that it has recently identified a build fail, then It can compute the diff between the two commits. That could be consider, to some extent, to the ground-truth patch. Even, for further scientific research, it would be interesting to collect X commits beyond the passing.

[monperrus]

very much related to @fermadeiral Bears' code

@dginelli has also done this for his paper, but manually.