ANN CNN backward propogation and draft docs

We now have working CNN and ANN implementation for backward pass
although it is quite tricky to use in the current state. There is still
missing a computational graph component to help link all the network
layers together rather than getting the user to do that themselves by
hand.

Documentation has many sections now populated and talks through some
subjects but its still not lets say a gentle introduction, it may need
some more hand holding as its pretty direct currentley.

Alpha

This version has a complete reseal, and re-array lib, with majority
support for neural networks, including forward pass.
The one thing lacking in this is backward propogation is still
experimental, and has many flaws that will need ironing out still.

Base ReSeal

Heavily worked around version, which allows serialisation and
deserialisation, with support for marshmallow schemas and validation.
Dockerisable, and unittest framework enabled for verification of
functionality. There are still some issues with automation, and size,
along with certain scenarios where it may be preferable to break the
routine, such as if you have added a plethora of ciphertexts together
you may still need to scale etc. But this should server as a basic
implementation with which to go forward.