Support memory and other uncompressive metric resource to use peak as histogram model data

[kitianFresh]

Describe the feature

1. now metric will be transformed to an internal signal, the signal use same process logic for any metric type, but for different type of metric, it has some different attribute, such as compressive & uncompressive resource.
2. pass some metric info to signal to do predict such as peak & oom value model update to make the real time predicting for evpa is more robust and safe