Also remove dead code in rist-common.c, it was being optimized away anyway.

The RIST committee in all it's wisdom decided to use some of the reserved bits for the version, instead of the existing GRE version field (likely to keep somewhat of compatibility with existing GRE implementations).

"RFC 8086 defines bits 4-12 in the GRE header as Reserved0.  This Specification defines bits 9-12 as follows (see Figure 1):
o   RV (bits 10-12): this field indicates the RIST Version, as follows:
§  RV = 000: TR-06-2:2020
§  RV = 001: TR-06 -2-2021
o   H (bit 9): this field indicates the AES key length for PSK operation, as follows:
§  H = 0: 128-bit AES key
§  H = 1: 256-bit AES key
·         Usage of the H bit shall be as follows:
o   If RV = 000, the H bit shall be set to zero on transmission and ignored on reception.  In this case, the PSK key length shall be determined by out-of-band means.
o   If RV = 001 or higher, the H bit indicates the PSK key length.  Devices not operating in PSK mode shall set the H bit to zero on transmission and ignore it on reception.
·         The remaining Reserved0 bits in Figure 1 shall be set to zero on transmission and ignored on reception, as per RFC 2784."

With this method, we are only importing the mbedcrypto library files and are
compiling them using meson/ninja direclty inside our main project.
This seems to solve all the cross-compiling issues and has no downside.
For the systems that already have the library installed, the compile
script will use that one instead by default.

Fixes #95 and #96

Fix peer and flow timeout checks. They were incorrectly being marked as dead when  the configured buffer was smaller than the RTCP intervals.

By setting the lowest 4 bytes of the IV to the sequence number IV's get reused.

Instead use set the top 4 bytes of the IV to the sequence number

This should prevent any blocking on retransmits on the output side

Fixes ffmpeg compilation in MinGW when linked with librist

Prevent crash from happening for the case of a peer connection coming from the same IP and same port as one of our listening peers

Fix mutex lock on peer list. This could potentially fix crashes related to peers being added or removed by applications after a flow has started.

EAI_NONAME is equal to 8, so instead of return there was null-pointer exception on res[0] access.
getaddrinfo returns 0 in case of success, thus need to check for non-zero result.

MacOS need's us to bind with INADDR_ANY & portno to respond back on multicasted packets, otherwise we get an error 49 when trying to send RTCP data back