gettext: derive charset from the environment when gettext is uninitialized

is_utf8_locale() drives, among other things, whether `git grep -P -i`
applies Unicode case-folding (grep.c sets PCRE2_UTF | PCRE2_UCP only when
this returns true).  It consults the static `charset` that
git_setup_gettext() fills in via locale_charset().  That setup, however,
bails out early when the locale directory cannot be found, leaving
`charset` NULL.  Because is_encoding_utf8(NULL) answers "yes", is_utf8_locale()
then reports a UTF-8 locale unconditionally, even under LC_ALL=C, and
`git grep -P -i` case-folds non-ASCII bytes when it must not.

I ran into this while running the test suite with BusyBox' ash as the
POSIX shell: in that leg the locale directory is not located, so t7816's
"LC_ALL='C' git grep -P -i" cases that assert no Unicode folding (the
matches_pcre2=0 rows) instead matched and their test_must_fail invocations
failed.  The MSYS2 Bash leg, where the locale directory is found and
locale_charset() returns "C", was unaffected.

The NO_GETTEXT build already protects against a NULL `charset` by deriving
it from LC_ALL/LC_CTYPE/LANG.  Apply that same fallback whenever `charset`
is NULL rather than only in NO_GETTEXT builds, so a gettext-enabled build
that could not initialize gettext still honors the locale from the
environment.

Assisted-by: Opus 4.8
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>

Author: dscho

gettext.c

@@ -141,7 +141,13 @@ int gettext_width(const char *s)
 
 int is_utf8_locale(void)
 {
-#ifdef NO_GETTEXT
+	/*
+	 * The charset is normally set by git_setup_gettext(), but it bails out
+	 * early when the locale directory is missing, leaving `charset` NULL.
+	 * As is_encoding_utf8(NULL) answers "yes", that would make us assume a
+	 * UTF-8 locale even under e.g. LC_ALL=C, so derive the charset from the
+	 * environment ourselves in that case.
+	 */
 	if (!charset) {
 		const char *env = getenv("LC_ALL");
 		if (!env || !*env)
@@ -154,6 +160,5 @@ int is_utf8_locale(void)
 			env = strchr(env, '.') + 1;
 		charset = xstrdup(env);
 	}
-#endif
 	return is_encoding_utf8(charset);
 }