Fix timestamps generation. Refactor the code concerning the mode change handling

[varsill]

No description provided.

[varsill]

Hi @gBillal ! Do you remember what was the reason for changing the prepare_timestamps implementation in this commit (while working on merging h264 and h265 parsers)?
It used to be Enum.find(au, &NALuTypes.is_vcl_nalu_type(&1.type)).timestamps and after this commit it is List.last(au).timestamps which causes problems when the last NAL unit in the AU does not have timestamps set and I wonder if I can safely revoke the old implementation.

[gBillal]

Hi @varsill
I thought that Since we only prepend parameter sets to the access units. It would be safe to assume that the last NAL unit of the AU will contain a correct timestamp (even if it's a NVCL NAL unit).

Seems that I'm wrong.

[varsill]

Sure, in most cases it works just fine, it's only problematic under some very specific circumstances, upon which we unfortunately came across ;)