RTP depacketizer: no STAP-A/AP, no CSRC/extension, no loss handling for video #10

NalAssembler only handles single NAL units and fragmentation units (FU-A 28 for H.264, FU 49 for H.265). Missing:

• Aggregation packets (STAP-A 24 for H.264, AP 48 for H.265) are passed to the decoder as one malformed NAL. Some servers send SPS/PPS this way.
• Video packets with CSRC or a header extension are dropped entirely ((data[0] & 0x0F) != 0 || (data[0] & 0x10) != 0). The audio path already accounts for both since fix(audio): AAC/G.711 playback; docs: README in EN/RU/ZH, device report form #6, via RtspClient.rtpHeaderLength().
• No sequence number tracking. A lost FU fragment produces a corrupted NAL, which is still handed to the decoder. A continuation fragment that arrives without a start fragment is appended to an empty buffer with no start code.
• NAL units larger than the buffer (1 MiB for single view, 512 KiB in quad) are silently truncated.

Proposal: parse the real header length, support STAP-A/AP, track RTP sequence numbers and drop the incomplete access unit on loss.