Float32 -> integer sample conversions (e.g. Float32_To_Int32()) needs to be documented and tested

[philburk]

We should document the wrapping behavior for the range +1.0 to -1.0.
And we should added tests to guarantee it on all platforms.
Wrapping would be very audible and a regression.

We are not currently documenting the linearity around zero.

[RossBencina]

The purpose of these tests is to ensure that any changes to the conversion functions (e.g. as discussed in #835) do not exhibit undesirable regressions. The worst type of regression would be if floating point values that currently convert without gross artifacts (e.g. wrap-around) suddenly start distorting.