if type checking gets too slow, try out moving away from fundeps

[cartazio]

currently, i do stuff like

class Array marr rank el | marr -> rank where 
   read :: Shape rank Int -> marr (PrimState m) el -> m el 
...

to pun naming, but fundeps have a less systematic theory/support in core ghc, and I'm told that compile (well type checking particularly) times can be dramatically improved by a moving to type families.

Should consider migrating to

class Array  marr el where
   type Ix marr :: Nat 
 -- might define Ix type family  out of band of the class 
-- so i can pun it between arrays and formats
   read :: (rnk ~ Ix marr) => Shape rank Int -> marr (PrimState m) el -> m el 

need to see how more complex algorithms wind up looking wrt types, and balancing usability/complexity