Cerberus 2: Coercion of nullable fields needs to (re)defined

[funkyfuture]

this issue is a reminder that the behaviour regarding the coercion (and broader normalization) of nullable fields needs to be thought through and defined for the next major release.

relevant information can be found at least in #427 and https://github.com/pyeve/cerberus/pull/490#discussion_r281262793.

#438 proposes a related, relevant change.

my general position is that Cerberus shouldn't try to be clever and imply one rule's behaviour due to the constraints of another, but give the user tools with a straight-forward behaviour that can handle every situation.

cc @andreymal, @dkellner and @nikhaldi. various user stories would be helpful here.