m4: use a safer pattern to enable/disable output

Work on some other areas of Bison revealed that some macros expanded
to be expanded only once were actually expanded several times.  This
was due to the fact that changecom was not properly restored each
time, and macro names appearing in comments were then expanded.

Introduce begin/end macros which are easier to match that
changecom()/changecom(#).

* data/bison.m4 (b4_output_begin, b4_output_end): New.
* data/glr.c, data/glr.cc, data/lalr1.cc, data/lalr1.java,
* data/location.cc, data/stack.hh, data/yacc.c:
Use them.

data/lalr1.java

@@ -22,8 +22,7 @@ m4_ifval(m4_defn([b4_symbol_destructors]),
         [b4_fatal([%s: %%destructor does not make sense in Java], [b4_skeleton])],
         [])
 
-m4_divert_push(0)dnl
-@output(b4_parser_file_name@)@
+b4_output_begin([b4_parser_file_name])
 b4_copyright([Skeleton implementation for Bison LALR(1) parsers in Java],
              [2007-2012])
 
@@ -925,4 +924,4 @@ b4_percent_code_get[]dnl
 }
 
 b4_epilogue
-m4_divert_pop(0)dnl
+b4_output_end()