TODO

* Testing framework!
    - Tasty batches
    - Test round trip of example .avro containers
    - Test round trip of each type.
        - Data.Avro level To/From Avro classes
        - Data.Avro.{Encode,Decode} level ToAvro/GetAvro classes
    - Test 'deconflict' for all pathalogical deconflications
* In-comment in-haddock tutorials and examples.
* Deal with 'order'?
* Refactor modules.
  - Schema also has lots of instances and parsing
  - 'Types' holds value types while shema hold type types - confusing!
* Make reasonable exceptions instead of using 'error'
* Lazily decode containers.
* Dog food some more: Container decoding is manual, use the damned schema!
* Handle namespaces properly
  - We don't account for identical schemas across name boundaries when resolving
    names (env)!
  - We pollute the env with all names unqualified, we should keep each env in a namespace
    and qualify the _query_.
* More 'FromAvro' and 'ToAvro' instances.
* More from/ToAvro helpers in the style of Aesons '.:'.
* Grep for XXX and fix - plenty of them.
* Remove dead code, commented out code and explorations.

Consider:
* Encoding unions in JSON is the reason we need the encoder schema... right? Crazy.
* Move to closer integration with Aeson
  * Move to using Aeson 'Value' instead of Avro.Types.Value?
  * Move to using Aeson 'Result'?  I want a MonadExcept instance.
* Use Attoparsec?  What would be the motivation away from `binary`?
* Make a composible schema deconfliction `deconflict :: Schema -> Schema ->
  Schema` ... requires embedding the decoding routine inside the schema so
  perhaps yuck.
* Prettier error messages, such as unresolvable schema's. (or, who cares,
  schemas can't be resolved deal with it?)
* Support RPC
* Make a .Generic module for encoding and decoding types that have generic
  derived.