Hi!

1. While I was working on helmfile + vals prototype as a part of helmfile#881
   I got an impression that given interface of vals don't use caching effectively enough: on each new Eval() call cache is recreated.
   So, what if we intorduce EvalRenderer object to vals api and each consecutive call of EvalRenderer.Eval() will use existing cache?

2. I see caching is implemented for each provider independently. What if we move caching one level up and decom caching for providers?

here is my propotype with proposed changes