netlink: add parseMessagesIter to reduce memory usage

[nickgarlis]

Introduce a parseMessages iterator to improve parsing performance and serve as the foundation for a future receive iterator in the public API.

The functionality in syscall.ParseNetlinkMessage is very similar to to Message.UnmarshalFromBinary; the only missing part was walking the buffer, which has now been implemented in this iterator.

Synthetic benchmarks show a performance improvement when parsing and collecting messages using this approach.

In more detail, the benchmark compared the old approach (syscall.ParseNetlinkMessage + conversion) with the new
iterator approach (parseMessagesIter + collection into slice).

goos: linux
goarch: amd64
pkg: github.com/mdlayher/netlink
cpu: 13th Gen Intel(R) Core(TM) i9-13900H
BenchmarkParseMessages/small/1msg/old-20         	 9447193	       117.5 ns/op	 136.17 MB/s	      96 B/op	       2 allocs/op
BenchmarkParseMessages/small/1msg/new-20         	15606260	        70.79 ns/op	 226.02 MB/s	      48 B/op	       1 allocs/op
BenchmarkParseMessages/small/10msgs/old-20       	  951582	      1174 ns/op	 136.23 MB/s	    1728 B/op	       6 allocs/op
BenchmarkParseMessages/small/10msgs/new-20       	 1230187	       915.4 ns/op	 174.79 MB/s	    1312 B/op	       5 allocs/op
BenchmarkParseMessages/medium/50msgs/old-20      	  221370	      5210 ns/op	 307.13 MB/s	    7840 B/op	       8 allocs/op
BenchmarkParseMessages/medium/50msgs/new-20      	  477787	      3678 ns/op	 434.96 MB/s	    5792 B/op	       7 allocs/op
BenchmarkParseMessages/large/100msgs/old-20      	  113692	      9720 ns/op	 823.07 MB/s	   16032 B/op	       9 allocs/op
BenchmarkParseMessages/large/100msgs/new-20      	  206064	      7015 ns/op	1140.45 MB/s	   11936 B/op	       8 allocs/op

[nickgarlis]

This is to keep the behavior similar to syscall.ParseNetlinkMessage

[nickgarlis]

cc @aojea in case you're interested

[nickgarlis]

BenchmarkNftablesDump shows about a 1.8% reduction in memory and a small drop in allocations. Nothing huge, but the larger improvements will come in the next change.

[aojea]

cc @aojea in case you're interested

thanks for doing this