Update to the current gnulib CVS repository, and fix trigraph handling

in Bison.
* bootstrap: Update gnulib CVS repository URL.
(symlink_to_dir): Encapsulate the code that guarantees the destination
directory exists into...
(check_dst_dir): ... this new function, and...
(cp_mark_as_generated): ... reuse it here so that bootstrap doesn't
fail when copying files into lib/uniwidth/.
* src/output.c (prepare_symbols): When writing yytname muscles, where
symbol names will be encoded in C-string literals, tell quotearg to
escape trigraphs.  This used to be the default in gnulib.
* tests/regression.at (Token definitions): Because of the change in
gnulib's quotearg behavior, string_as_id in parse-gram.y no longer
escapes trigraphs in symbol names.  Thus, yytname no longer has
trigraphs unnecessarily doubly escaped.  Update test case output.
Extend test case to be sure Bison's own error messages will no longer
have trigraphs in symbol names unnecessarily escaped once.

lib/.cvsignore

@@ -25,6 +25,7 @@ fd-safer.c
 fopen-safer.c
 getopt.c
 getopt.h
+getopt.in.h
 getopt1.c
 getopt_.h
 getopt_int.h
@@ -32,7 +33,9 @@ gettext.h
 gnulib.mk
 hash.c
 hash.h
+intprops.h
 inttypes.h
+inttypes.in.h
 inttypes_.h
 localcharset.c
 localcharset.h
@@ -52,17 +55,21 @@ ref-del.sed
 ref-del.sin
 stamp-h1
 stdbool.h
+stdbool.in.h
 stdbool_.h
 stdint.h
+stdint.in.h
 stdint_.h
 stdio--.h
 stdio-safer.h
 stdlib.h
+stdlib.in.h
 stdlib_.h
 stpcpy.c
 streq.h
 strerror.c
 string.h
+string.in.h
 string_.h
 stripslash.c
 strndup.c
@@ -74,6 +81,7 @@ strverscmp.h
 unistd--.h
 unistd-safer.h
 unistd.h
+unistd.in.h
 unistd_.h
 unitypes.h
 uniwidth
@@ -81,8 +89,10 @@ uniwidth.h
 unlocked-io.h
 verify.h
 wchar.h
+wchar.in.h
 wchar_.h
 wctype.h
+wctype.in.h
 wctype_.h
 wcwidth.c
 wcwidth.h