[0.x] Optimize NACKs handling

[atoppi]

This PR includes a set of optimizations for NACKs handling.
We identified some inefficient code paths that could potentially cause some performance issues when the server receives a high number of NACK requests.
Once completed this, a similar effort will be done on 1.x too.

TODO

• replace g_list with g_queue to only use O(1) operations and avoid full list scanning
• use a pre-allocated queue per component
  reduce memory allocations when preparing packets to retransmit
• add a backoff algo when retransmitting the same packet multiple times

[atoppi]

This is now ready to be tested.

[atoppi]

Reverted all allocation-related stuff since it was not a viable solution.