Separate determination of GCC_build for host and target

[giordano]

The choose_shards function is rather complicated. Among its issues it computes the GCC_build for the host and the target together, which for example fails for

BinaryBuilderBase.runshell(Platform("x86_64", "linux"; libc="musl", cxxstring_abi="cxx03");preferred_gcc_version=v"4")

GCC v4.8 is chosen for both of them, when it should choose v4.8 for the target and v5 for the host (with #82 we always require C++11 string ABI for the host). The same happens for the target Platform("x86_64", "linux"; libc="glibc", cxxstring_abi="cxx03") and I have an even harder time to understand why, since the two platforms are strictly speaking different (maybe the same version of GCC is always chosen?). See also #82 (comment), which is what we want to eventually achieve