Enable link-time optimizations for release builds to reduce the binary size

[rdw-software]

This reduces the binary size by ~7 MB on Windows, but adds several seconds to each rebuild. Since there's currently just a single build configuration, adding this in a non-hacky way would require #128. Always enabling it increases iteration time (bad idea).

Also, at least the MSYS2 gcc defaults to using a single core when no argument is given (instead of defaulting to use the maximum), which means something like nproc has to be used to compute the maximum at configuration time.

[rdw-software]

Seems to be blocked because of symbol type issues with uSockets. Don't have time to investigate, but might be Windows only?

Edit: See WITH_LTO setting and comment. I don't remember facing this previously; misconfiguration somewhere?