Fix "ip broadcast" netmask byte order with the -f flag.

Fix "ip broadcast" netmask byte order with the -f flag.

Let's suppose the interface eth0 has one IPv4 address with a /24
netmask.  Without -f tcpdump leaves the netmask variable set to 0, which
regardless of the host endianness causes "ip broadcast" to match
destination hosts 0.0.0.0 and 255.255.255.255:

# tcpdump -i eth0 -d 'ip broadcast'
(000) ldh      [12]
(001) jeq      #0x800           jt 2	jf 6
(002) ld       [30]
(003) jeq      #0x0             jt 5	jf 4
(004) jeq      #0xffffffff      jt 5	jf 6
(005) ret      #262144
(006) ret      #0

With -f tcpdump calls pcap_lookupnet(), which correctly sets the netmask
to 0xFFFFFF00 (in network byte order).  Then pcap_compile() receives the
same value, but it expects it to be in host byte order, so on a
little-endian host the resulting filter program incorrectly tests for a
0x00FFFFFF netmask:

# tcpdump -i eth0 -f -d 'ip broadcast'
(000) ldh      [12]
(001) jeq      #0x800           jt 2	jf 7
(002) ld       [30]
(003) jset     #0xff000000      jt 4	jf 6
(004) and      #0xff000000
(005) jeq      #0xff000000      jt 6	jf 7
(006) ret      #262144
(007) ret      #0

Add two missing ntohl() wrappers to make it right:
# tcpdump -i eno1 -f -d 'ip broadcast'
(000) ldh      [12]
(001) jeq      #0x800           jt 2	jf 7
(002) ld       [30]
(003) jset     #0xff            jt 4	jf 6
(004) and      #0xff
(005) jeq      #0xff            jt 6	jf 7
(006) ret      #262144
(007) ret      #0

Audit the init_print() code path and do not change anything because
there the byte order is already correct.  Add comments to spell the byte
order in every case and update the -f flag description in the man page.
See also libpcap commit 1e54958.