Run parallel inference test on each accelerator

Add to the parallel inference test currently running without setting
the target accelerator name a series of tests with same duration and
number of threads but targeting each available accelerator.

Add support for target accelerator selection to the test app.

Test: build_and_run_benchmark.sh parallel-inference-stress
Test: run  NeuralNetworksApiCrashTest app on device
Change-Id: Ib945280ecb4585c5466c45907cd5c0524842cc76
5 files changed