Now that we gonna have soft device affinity that allows swapping out failed perf devices ( issue 831252 ), we should track how often devices get swapped out since this affect the quality of perf monitoring service.

A way to implement this is using perf dashboard. The reason is we gonna upload the perf device id to the perf dashboard anyway for detecting false regressions due to perf device swap.

An example simple "device affinity" metric:
Σ P(test, build N, build N +1) for all perf test

whereas:
P(test, build N, build N +1) = 1 if test run on same device between build N & N+1, 0 otherwise.

File this bug to track so someone can take it later.