Allow both sender and receiver to generate packet loss on their outgoing packets, to be used in testing.

Also loop over children of parent rtcp peer instead of global peer list & remove assert

(Park Run 1280x1080)

For now we just output the data stats, RTCP stats can be fixed later

Fix packets not being recovered in simple profile as well as crashing under higher packet loss

Allow a null value for the stats callback pointer to allow setting the statsinterval to 0 without having to pass a valid pointer

Disable advanced profile for this release as it is not in the spec yet and it has not been properly tested (fixes issue #30)

Set default statstimer for receiver context and do not set callback object when statstimer is zero (fixes issue #28).

Unrecognized rtcp packet messages moved down from warn to debug, was spamming logs.
Also added XR recognition with debug message stating it's not handles (perhaps we can handle it in the future, at least one implementation is using them).

With this commit, the tools now have two modes of operation.

Mode 1 input: you use udp:// as input url and the payload will not be touched
Mode 2 input: you use rtp:// as input url and the payload will be stripped of the rtp header

Mode 1 output: you use udp:// as output url and the payload will be an exact copy of the input from the sender side
Mode 2 output: you use rtp:// as output url and the ristreceiver will append an rtp header to the payload before outputting it to the network

You are allowed to use any combination:

udp:// input and udp:// output (old and only option prior to this commit)
udp:// input and rtp:// output
rtp:// input and udp:// output
rtp:// input and rtp:// output

Use a separate config structure and parser for udp url (this gets rid of the ugly hack of using the peer config parser)

Change the default behavior to perform transfers without using multiplexing/demultiplexing rules (it makes it more friendly to non-librist peers)

This was causing the RTP streams to have a virtual destination port of 0. librist receivers did not care as they do not check it. However, it was making our stream out of compliance and causing problems with netinsight receivers.

Fix bug where the session expiration on receiver was causing the master listening peer to not process any more data events.

This would basically leave the receiver with an open socket but in a stuck state (new connections would be ignored).

Switch all recvfrom and sendto socket calls to be non-blocking for higher performance and reliability (poll will still block and maintain timing)

Fix regression on socket event management that lead to deadlocks (because of blocking recvfrom use) when multiple peers listeners where used and also broke the handshake of simple profile connections

Defining htobe64/be64toh if below a minimum glibc version does not work
under musl libc where __GLIBC__ etc are not defined.

Rather than providing another logic branch for the musl case, just test
directly to see if the definitions are present.

When I try to compile against musl libc on Ubuntu 20.04, linux/if_alg.h
is not available.

This change simply uses the fallback code if the header isn't found at
configure-time.

I'm not sure if there's some missing package that would provide this.  I
searched but did not spot a candidate.