Fix repo diff when diff=cpp attribute is used with utf-8 locales (on macOs) - #7399

With *.c diff=cpp, generating an ordinary patch can fail on macOS with
"illegal byte sequence" under a UTF-8 locale. The built-in word pattern
contains this C string fragment:

    "[\xc0-\xff][\x80-\xbf]+"

The C compiler turns these escapes into raw bytes that are not valid UTF-8
in the pattern, so regcomp_l rejects it while loading the driver.

The word patterns were imported from Git alongside the function name
rules. The compiled word_pattern is stored and freed, but never used for
matching. Hunk headers use the separate fn_patterns list, so omitting
word-pattern compilation does not change hunk contents or headers.

Commit ab96ca5 ("Make sure we use the `C` locale for `regcomp` on
macOS.") addressed this error by switching to regcomp_l, but the failure
is still reproducible under a UTF-8 locale on macOS.

Stop compiling built-in word patterns when loading a driver and add a
regression test for diff=cpp under a UTF-8 locale.

Reproduce on macOS with pygit2 linked to libgit2 using regcomp_l:

python3 - <<'PYTHON'
import locale
from pathlib import Path
from tempfile import TemporaryDirectory
import pygit2

locale.setlocale(locale.LC_CTYPE, 'UTF-8')
with TemporaryDirectory() as directory:
    repo = pygit2.init_repository(directory)
    root = Path(directory)
    (root / '.gitattributes').write_text('*.c diff=cpp\n')
    source = root / 'test.c'
    source.write_text('int value = 1;\n')
    repo.index.add('test.c')
    repo.index.write()
    source.write_text('int value = 20;\n')
    print(repo.diff().patch)
PYTHON

Before the fix, generating the patch raises "illegal byte sequence".

Signed-off-by: Gregory Herrero <gregory.herrero@oracle.com>