Add fixed-length floating point number support

Add fixed-length floating point number support

This patch adds support to encode binary32 and binary64 (as per
IEEE 754-2008) floating point numbers, big and little endian.

The Normand form is the same as a fixed-length integer:

    { 32.87 : 32 }

The type of the result of the evaluation determines how to encode it:
integer or floating point number.

Some "fixed-length integer" terms are changed to "fixed-length number"
when they're meant to be general.

With this patch, you may also assign a variable to a floating point
number value. However, the result type of an LEB128 integer or
repetition count expression must still be `int`.

Change-Id: Ic13ed747276a365ed2fd8854545ab2b0e1c343a0
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>