Fix "float unrepresentable in integer range" runtime error due to incorrect unsigned conversion

[TerrorJack]

Closes #629. Thanks to @mikesol @InBetweenNames, we located a very old runtime bug. The symptom is a "float unrepresentable in integer range" runtime error, and the root cause is:

• We have quite a few rts builtin functions which are implemented in JavaScript, and imported into WebAssembly
• Since our word size is 64-bit and our WebAssembly functions mostly use i64 in parameter/return types, the JavaScript functions need to be imported to use f64, and a layer of wrapper functions will generate casts between i64 and f64.
• Our mechanism to generate the wrapper functions incorrectly used an unsigned cast for function results. When an rts function returns a negative result (e.g. _hs_text_memcmp), this will result in an error.

In the longer term we should be using the linear memory to pass FFI paramters/results in a uniform manner, without worrying about precision losses or signedness issues. But this PR suffices to fix the error for now.

[InBetweenNames]

Happy to help!