track currently active features while scanning

[mauke]

... instead of repeatedly scanning backwards for all feature enabling/disabling statements every time we need to know the current feature set. In my tests, this massively speeds up parsing of big documents/source files.

Fixes #309.

Demonstration:

use v5.36;
use PPI::Document;

my $source = "use v5.36;\n";

for my $i (0 .. 399) {
    my $sub = "sub foo_$i(\$x, \$y) {\n";
    for my $j (0 .. 29) {
        $sub .= "    g(\$x->[\$y]" . join('', map ", $_", 1 .. $j) . ");\n";
    }
    $sub .= "}\n";
    $source .= "$sub\n";
}

say "parsing " . length($source) . " chars of code ...";
my $doc = PPI::Document->new(\$source);
say "done";

With PPI 1.284:

$ time perl try.pl
parsing 819901 chars of code ...
done

real    0m37.017s
user    0m36.700s
sys     0m0.310s

With this patch:

$ time perl -I"$HOME/Projects/PPI/lib" try.pl
parsing 819901 chars of code ...
done

real    0m7.387s
user    0m7.170s
sys     0m0.217s