Compile error instead of falling back to C for unsupported calling conventions

[tadeokondrak]

Closes #6536.

[tadeokondrak]

Trying to build the self-hosted compiler for x86_64-windows-gnu on my machine seems to cause lld to call _exit somewhere during linking. Not sure why.

[LemonBoy]

Suggested change

	allowed_platforms = "x86, x86_64, AVR, and MS430";
	allowed_platforms = "x86, x86_64, AVR, and MSP430";

[LemonBoy]

The old behaviour was modeled after clang as it ignores (and afair gives a warning) whenever it encounters a invalid cc for the given target. But giving an error is also fine and will force people to use the correct cc, so +1 from me.

[Rocknest]

How do i use SysV callconv when target os = windows?

[tadeokondrak]

How do i use SysV callconv when target os = windows?

I don't think this is possible but it should be. I don't think it's related to this PR though.

[tadeokondrak]

Still don't know why this is failing, so closing until #3825 is fixed

[LemonBoy]

Still don't know why this is failing, so closing until #3825 is fixed

You should be able to see LLD's fatal errors by modifying ZigLLDLink to use llvm::errs() and llvm::outs() instead of diag_stderr and diag_stdout, the former are two unbuffered streams going to the terminal while the latter are the fake ones Zig setups in order to collect the error messages.
When LLD calls fatal() the whole Zig process is terminated and the error messages get stuck in the diag streams.