traces: provide a means to get short m4 traces

Let --trace=m4-early dump all the logs from the start (as --trace=m4
used to do), and have --trace=m4 now start traces only when actually
working of the user's grammar.

Can make a big difference in the case of small inputs.  E.g.

    $ bison -S tests/testsuite.dir/001/input.m4 tests/testsuite.dir/001/input.y --trace=m4 |& wc
       3952   19446  251068
    $ bison -S tests/testsuite.dir/001/input.m4 tests/testsuite.dir/001/input.y --trace=m4-early |& wc
      19491  131904 1830495

* data/skeletons/traceon.m4: New.
* src/getargs.h, src/getargs.c: Introduce --trace=m4-early.
* src/output.c (output_skeleton): Adjust for --trace=m4 and --trace=m4-early.

src/getargs.c

@@ -257,7 +257,8 @@ static const argmatch_trace_doc argmatch_trace_docs[] =
   { "sets",       "grammar sets: firsts, nullable etc." },
   { "muscles",    "m4 definitions passed to the skeleton" },
   { "tools",      "m4 invocation" },
-  { "m4",         "m4 traces" },
+  { "m4-early",   "m4 traces starting from the start" },
+  { "m4",         "m4 traces starting from the skeleton evaluation" },
   { "skeleton",   "skeleton postprocessing" },
   { "time",       "time consumption" },
   { "ielr",       "IELR conversion" },
@@ -279,6 +280,7 @@ static const argmatch_trace_arg argmatch_trace_args[] =
   { "sets",      trace_sets },
   { "muscles",   trace_muscles },
   { "tools",     trace_tools },
+  { "m4-early",  trace_m4_early },
   { "m4",        trace_m4 },
   { "skeleton",  trace_skeleton },
   { "time",      trace_time },