Re: Ensure hashable are Serializable

[Chris Warburton]

Turns out we had a bug in setting 'hashable'.

We try a hashed version and an unhashed version, and see what works. For
each function name, we see if anything with that name worked, and if any
of those was a hashed version.

We forgot this second check, so non-hashable things were being flagged
as hashable, simply because we found some working output (despite it
being from the unhashed version).

This is now fixed, with regression tests involving dlist: `replicate`
produces a `DList a`, which synonymous with a function type and hence is
not hashable; `toList` produces a `[a]` which is hashable.