Implement per-platform compiler version selection

[imciner2]

This implements the ability to select a different compiler version for each requested platform. This can be useful when certain platforms might need a newer version of GCC to fix bugs in the toolchain (e.g., riscv), but we don't want to increase the version for working platforms.

An example use of this is:

platforms = supported_platforms()
platforms = expand_cxxstring_abis(platforms)

# Use GCC 13.2 on RISC-V to fix linking issues with atomic operations, but GCC 6 everywhere else
compiler_versions = preferred_platform_compiler(platforms, v"6")
set_preferred_compiler_version!(compiler_versions, v"13.2", p -> arch(p) == "riscv64")

build_tarballs(
    ARGS,
    name,
    version,
    sources,
    script,
    platforms,
    products,
    dependencies;
    preferred_gcc_version = compiler_versions,
    julia_compat = "1.6",
)

This will then use GCC 13 on the RISCV platforms, and GCC 6 on all other platforms.

[ararslan]

I wonder if it would be simpler, both in terms of implementation and user interface, to do something like, for example:

# In the Yggdrasil `build_tarballs.jl` script:
build_tarballs(ARGS, name, version, sources, script, platforms, products, dependencies;
               preferred_gcc_version=(p -> Sys.islinux(p) && arch(p) == "riscv64" ? v"14" : nothing))

When preferred_gcc_version is a function instead of a particular version number, the version is selected by applying the function to each element of the provided array of platforms, with nothing serving to denote the default value. The same behavior could be applied to the other compiler preference keyword arguments as well.