RFE: A version of Inference[P, C] that fails for implicit chaining

I could use a version of `Inference` that always fails to manifest. For example, the following inference "fails" but it doesn't fail to manifest, it just has a `false` value:
```scala
scala> implicitly[Inference[Greater[-1], Greater[0]]]
res7: eu.timepit.refined.api.Inference[eu.timepit.refined.numeric.Greater[-1],eu.timepit.refined.numeric.Greater[0]] = Inference(false,greaterInference(-1, 0))
```
This works for the casting in `InferMacro` and `autoInfer` but it's limited in how it can be used in chained implicits.

I could use an alternative that will fail like so:
```scala
scala> implicitly[ChainableInference[Greater[-1], Greater[0]]]
error: could not find implicit value ...
```
Given the current design, I think that another similar macro which returns a chainable inference type but only if `inference.notValid` is false, instead of just doing an implicit value conversion.