[clang] Fix encoding handling of asm strings (#169803)
This patch builds upon https://github.com/llvm/llvm-project/pull/138895
and introduces a new ConversionAction parameter to
ParseStringLiteralExpression to allow asm strings to be, for z/OS,
initially evaluated in IBM-1047 (so that numeric escape sequences denote
IBM-1047 code points) before being converted to UTF-8 for further
processing (e.g., parsing for operand references).
For example, on z/OS we want \174 to map to the '@' character in
IBM-1047 regardless of fexec-charset.
`__asm__("\174asm_sym")` This is required because there are variant
characters in the ebcdic family and it is common on the z/OS platform to
compile with an fexec-charset set to an ebcdic (non-IBM-1047) codepage.
This would result in different values for ("@asm_sym")
If there is a constexpr in place of the asm string, the sequence of char
values produced will be converted (for all targets, not only for z/OS)
from the literal encoding to UTF-8 for further processing, except when
the literal encoding is UTF-8.