These predefined macros indicates that our u and U chars/strings
are UTF-16 and UTF-32-encoded, respectively.

On most Unix-like systems, wide character literal is 32-bit long
and encodes a Unicode code point. On Windows, that is 16-bit
long and encodes a UTF-16 code unit. Clearly, there's a portability
issue here. Personally I've never used wide characters in my code
as I didn't find it useful.

Being said that, some header files contain wide character literal,
so we need to support that so that chibicc can include such files.

We assume that source files are always encoded in UTF-8.

Quote from AMD64 System V ABI: "An array uses the same alignment as
its elements, except that a local or global array variable of length
at least 16 bytes or a C99 variable-length array variable always has
alignment of at least 16 bytes."

We don't want to leave a partial assembly output if the compiler
fails during compilation.

Technically speaking, there's still a risk of leaving a partially-
written output file if the compiler dies during file copy. To fix
that, we have to write to a temporary file in the same filesystem as
the final output file and rename the temporary file to atomically
replace the output file. We don't do that in this patch for the sake
of succinctness, though.