Add `deepseq` on the input for `nf` and `whnf`

This is to ensure that the input evaluation is not involved in the execution of the benchmark.

Here is the patch that I'm suggesting:

```diff
-nf :: NFData b => (a -> b) -> a -> Benchmarkable
-nf f x = toBenchmarkable (nf' rnf f x)
+nf :: (NFData a, NFData b) => (a -> b) -> a -> Benchmarkable
+nf f x = deepseq x (toBenchmarkable (nf' rnf f x))
```