Fixed point decimal representation.

[cgyurgyik]

Testing my recent changes a bit more thoroughly, I believe there are two ways this can go.

(1) The user passes the following:

{
  ... 
  data: [ 0.1 ]
}

The value pulled from the JSON key data is now: 0.1000000000000000055511151231257827021181583404541015625. This value is indeed representable by fixed point, given enough fractional bits, so the error they receive is fractional width overflow, required fractional width: 55..

(2) The user passes in the data as strings, e.g.

{
  ... 
  data: [ "0.1" ]
}

This ensures that 0.1 indeed has the value 0.1, and thus the error they receive is not representable by fixed point numbers: 0.1.

To me, (2) seems less surprising to the user, and we can easily ensure that strings are actually passed for fixed point. Option (3) might be to explore other JSON libraries for Python that gives us more control over how numbers are interpreted.

[cgyurgyik]

I'm going to spend some time boxing up Fixed Point number and Bitnum number representations in respective Python classes to make it a bit more clean and extensible. For simplicity, I'll just assume you pass in a string version of the number to the class.

[cgyurgyik]

Looks like simplejson provides this option. We can avoid the cast to float and directly convert it to a string instead.