Some test portability fixes again + minor changes

bench.PL

@@ -223,6 +223,7 @@ sub fmtbench($id, $text, $xs, $ys) {
     open my $F, '>FU/Benchmarks.pod' or die $!;
     select $F;
     while (<DATA>) {
+        s/^%/=/;
         s#^:modules#join '', map sprintf("=item L<%s> %s\n\n", $_, $modules{$_}), sort keys %modules#e;
         s#^:benches (.+)#join '', map fmtbench(@$_), grep $_->[0] =~ /$1/, @bench#e;
         print;
@@ -233,12 +234,13 @@ sub fmtbench($id, $text, $xs, $ys) {
     }
 }
 
+# s/^=/%/  to prevent tools from interpreting the below as POD
 __DATA__
-=head1 NAME
+%head1 NAME
 
 FU::Benchmarks - A bunch of automated benchmark results.
 
-=head1 DESCRIPTION
+%head1 DESCRIPTION
 
 This file is automatically generated from 'bench.PL' in the L<FU> distribution.
 These benchmarks compare performance of some FU functionality against similar
@@ -256,19 +258,19 @@ real-world use.
 B<DISCLAIMER#3:> Many of these benchmarks exists solely to test edge case
 performance, these numbers are not representative for real-world use.
 
-=head1 MODULE VERSIONS
+%head1 MODULE VERSIONS
 
 The following module versions were used:
 
-=over
+%over
 
 :modules
 
-=back
+%back
 
-=head1 BENCHMARKS
+%head1 BENCHMARKS
 
-=head2 JSON Parsing & Formatting
+%head2 JSON Parsing & Formatting
 
 These benchmarks run on large-ish arrays with repeated values. JSON encoding is
 sufficiently fast that Perl function calling overhead tends to dominate for
@@ -279,10 +281,10 @@ SIMD parts are only used for parsing.
 
 :benches ^json
 
-=head2 XML Writing
+%head2 XML Writing
 
 :benches ^xml
 
-=cut
+%cut
 
 # Cached data used by bench.PL. Same as the formatted tables above but easier to parse.