Add timestamp generation for profiles baseline, constrained baseline

[sgfn]

Timestamp generation for profiles which may contain B-frames not yet implemented.

[varsill]

Great job! I am really impressed by the quality of the code you have provided ;)
My only concern is about the prepare_actions_for_aus function - I think we are mixing up the terms "timestamp" and "ordering_number" (in case of both maybe_generate_timestamps and state.previous_timestamps). Timestamps might be linearly correlated with the ordering_numbers in case of the constant framerate, but we shouldn't use these terms interchangeably.
I would suggest the following changes:

1. Add au_counter to the state, which would hold the number of access_units (frames) already processed
2. Simplify prepare_actions_for_aus function - it could look like that:

   {actions, au_counter, profile} =
      Enum.reduce(aus, {[], state.au_counter, state.profile}, fn au, {actions_acc, cnt, profile} ->
        {sps_actions, profile} = maybe_parse_sps(au, state, profile)
        {pts, dts} = prepare_timestamp(buffer_pts, buffer_dts, state, profile, cnt)
        {actions_acc ++ sps_actions ++ [{:buffer, {:output, wrap_into_buffer(au, pts, dts)}}], cnt + 1,
         profile}
      end)

    state =
      maybe_update_state(buffer_pts, buffer_dts, au_count, profile, state)
    {actions, state}

3. prepare_timestamp would hold the whole logic of creating/rewritting of the timestamps (perhaps it might be worth to provide three different implementations, based on the state.mode)

[sgfn]

@varsill Thanks for the review and the kind words :)) hope to have addressed all of your concerns with these fixes

[varsill]

Yeah, that's what I have meant ;)
There are some NITs you can consider implementing and one issue concerning the functionality - I think with the parser working in :nalu_aligned mode the first frame would be send with pts: nil and dts: nil

[sgfn]

Good catch -- fixed. Thanks

[varsill]

LGTM! 🥇