added float32 option

[akotlar]

Creates new flag, prefer_float32, which allows the use of 5 bytes per float, instead of the default 9.

Flag is named "prefer_float32" to anticipate this library's move to more recent versions of msgpack-c, which rename float/double functions to float32/float64.

For population genetics computations, 32bit float is quite enough.

Addresses #31