Link chpl with the same libc++ that built the Homebrew LLVM

[mppf]

Resolves #19217

This PR adds a workaround to solve #19217. The problem we were seeing is that the Homebrew LLVM is built with one C++ standard library, but it includes a different version of a C++ standard library that our include and link paths were picking up. This can be detected with comparing output of otool -L chpl vs otool -l libLLVM.dylib (with appropriate paths).

To fix it, this PR adjusts chpl_llvm.py to:

• compute the sdk used when building the llvm/clang Homebrew package. The current strategy is to find the DEFAULT_SYSROOT define from clang/Config/config.h. This strategy relies on the Homebrew formula passing -DDEFAULT_SYSROOT=#{macos_sdk} when cmake'ing llvm and clang. The main alternative strategy that I know of for this is more complex and relies on running both otool and xcrun.
• Put -I and -L paths based on this sdk first in the system compile options so that we find the sdk libc++ rather than one from the Homebrew package

While there, I fixed some problems in error reporting when checking an llvm-config.

Reviewed by @ronawho - thanks!

Testing:

• thanks to @ronawho and @aconsroe-hpe for helping test some of these configs
• full local testing
• Ubuntu 21.10 system still works with system LLVM
• Ubuntu 21.10 system with homebrew works with homebrew llvm@12
• Monterey (12.2) ARM Mac with llvm@11, llvm@12, and llvm@13
• Monterey (12.2) x86 Mac with llvm@11, llvm@12, and llvm@13
• Big Sur (11.6) x86 Mac with llvm@12 and llvm@13
• Catalina (10.15.7) x86 Mac with llvm@11, llvm@12, and llvm@13
• some of the above Mac systems have xcode and some only have command line tools