Generic tries (long)

[Jamie Brandon]

Ive added the rest of Adrian's functions to the api, largely without
change. Still living at http://code.haskell.org/gmap/api/GMap.html

The Serial code is now *finished* and tested and is living at
http://code.haskell.org/gmap/serial . The user level interface is
largely the same as Data.Binary, on which it is based:

*Data.Serial> encode "supercalifragilisticexpialidociousness" :: [Word]
[1401017574,815320646,2442711660,441247251,3504761142,212218697,3261675550,3863169456,1865717965,1399966954,6082150]

I'll get to work on some benchmark now so I can see how efficiently
the simple version of the api can be implemented. It doesnt really
seem worth arguing over it until we have some numbers. I would hope
that if merge and the associated comparison functions are declared
inlinable then ghc should be able to mostly recover the original,
efficient implementations.

Cheers

Jamie