Edge AI Benchmark: Consolidate/Prioritize metrics

[YuxiaoHu]

The draft benchmark version looks good, which is a nice start point. Thanks for the great efforts!

Issues: There are too many metrics, which lead to two problems:

• Difficult for readers to understand each of them according to the short descriptions. This proposal should be self-contained, so that the readers won’t need to read all the references. It will be nice to add descriptions/examples/Figures for each of the metrics;
• For new users, it will be difficult to execute. I suggest we start from one or two core metrics for each scenario (like P0 features), and provide additional metrics (as P1/P2, …) for advanced users or readers who are interested.

[MooreZheng]

Thanks. We have updated the first draft of the specification.

The specification is still open to edit. Any interested member please also refer to #96

[MooreZheng]

Add as suggested, e.g., https://github.com/kubeedge/ianvs/tree/main/examples/pcb-aoi/singletask_learning_bench/testenv