Add sub signatures performance note to perldelta

[bbrtj]

This adds a note to perldelta about signatures having less runtime overhead due to internal changes made to support named parameters (or so I heard).

This was discussed on IRC, and I checked that there is indeed a performance difference. Here is the benchmark code:

use v5.42;

use Benchmark qw(cmpthese);

sub s_raw { my ($x1, $x2, $x3, $x4) = @_; }
sub s_sig ($x1, $x2, $x3, $x4) {}

cmpthese -2, {
	raw => sub { s_raw($_, $_, $_, $_) for 1 .. 1000 },
	signatures => sub { s_sig($_, $_, $_, $_) for 1 .. 1000 },
};

It consistently shows raw case being ~25% faster than signatures on 5.42, and ~5% faster on compiled blead. While this is not the best way to benchmark, that much of a difference should be enough to determine that indeed signatures are now more efficient. Note that raw only serves as a base for comparison here, so that we see efficiency difference between two variants of argument unpacking, without the need to take overall perl interpreter efficiency into account.

Example results:

$ ~/localperl/bin/perl bench.pl
             Rate signatures        raw
signatures 7378/s         --        -3%
raw        7622/s         3%         --

$ perl bench.pl
             Rate signatures        raw
signatures 5183/s         --       -25%
raw        6950/s        34%         --

---

• This set of changes requires a perldelta entry, and it is included.

[leonerd]

Also while you're there, all of the entries should be their own =item lines within a single =over / =back pair, rather than have their own ones.