Fix DECnet packet filtering on big-endian hosts.

Fix DECnet packet filtering on big-endian hosts.

DECnet address matching works on little-endian hosts only:

$ arch
OpenBSD.amd64
$ tcpdump -nr tests/DECnet_Phone.pcap -c 1 'decnet src 1.1' | wc -l
reading from file tests/DECnet_Phone.pcap, link-type EN10MB (Ethernet),
  snapshot length 65535
       1

$ arch
OpenBSD.mips64
$ tcpdump -nr tests/DECnet_Phone.pcap -c 1 'decnet src 1.1' | wc -l
reading from file tests/DECnet_Phone.pcap, link-type EN10MB (Ethernet),
  snapshot length 65535
       0

Although the savefile is the same, the bytecode is different: in the
latter case gen_dnhostop() generates 8 statements out of 20 with 16-bit
literal values byte-swapped, e.g.:

-(006) jeq      #0x104           jt 22	jf 7
+(006) jeq      #0x401           jt 22	jf 7

For one half of those statements the root cause is feeding the provided
16-bit big-endian DECnet address, which always needs to be converted to
little-endian to match the packet encoding, to ntohs(), which leaves the
value intact on a big-endian host.  Fix this by using SWAPSHORT()
instead.

For the other half the root cause is treating two adjacent independent
bytes of the packet as a 16-bit value, hard-coding it and its 16-bit
mask byte-swapped and feeding these values to ntohs(), which on a
big-endian host does not cancel the hard-coded byte swap out.  Fix this
by hard-coding the two bytes and the bit mask as big-endian in the first
place.

Remove a number of unnecessary type casts and add a comment to explain
the packet encoding and the code logic.