Reimplement embed-bitcode part as a separate pass

Invoke -flto=thin in the first stage of -ffat-lto-objects

Add a clang Driver test

Use obj2yaml and yaml2obj to avoid unreadable object files

Replace section name from .fatlto to .llvm.lto

Add clang and llvm related changes

In D131618#3727832, @tstellar wrote:

Is there a clang part of this change?

Add a draft test. Not sure if this is the best way to verify the executables
are identical. I should probably ignore sections like .comment. I would
appreciate a feedback on the approach.

Check for the pointer validity separately with early continue.

Check for the pointer validity separately

Do not aggregate the .fatlto section when generating non-LTO executable.

Add the -fat-lto-objects flag as a prerequisite to checks.

Fix the order of bitcode and fatLTO section check.

Reverting the magic number based design. FatLTO object files stay as ELF files
with embedded bitcode. That's also how GCC does it. Current change checks if
there is a bitcode section named .fatlto.