avtp: Various memory and socket handling issues

The following discussions from !361 (merged) should be addressed:

• @slomo started a discussion:

  This is suboptimal as it blocks until a packet is received, which makes pipeline shutdown a bit exciting :) This should probably make use of GstPoll or directly use the gio GSocket API. Unblocking this function here should happen as part of GstBaseSrc::unlock.

• @slomo started a discussion:

  Casting raw data like this is inviting SIGBUS and endianess issues and all kinds of undefined behaviour. It would be better to use a GstByteReader or similar here.

  The same probably applies to other code too, but this one caught my eye.

• @slomo started a discussion:

  Raw audio data needs to be properly aligned according to the basic sample type (e.g. 2-byte aligned for S16). This is not ensuring that the alignment is correct.

That's what I found while looking for 5 minutes. The whole plugin should probably get another review.