Prevent delayed FrameDecoded from advancing RTP cleanup epoch

FrameDecoded() unwraps a stored uint16_t RTP sequence number relative
to the current receive state. If the frame is decoded only after the
receive sequence advances by more than half the 16-bit range, the old
value can unwrap into the next epoch. PacketBuffer::ClearTo() then
advances the cleanup boundary into that future epoch, and subsequent
packets from the current epoch are dropped.

Use PeekUnwrap() so delayed decode feedback cannot modify the receive
unwrapper. If the resulting cleanup boundary is ahead of
newest_media_seq_num_, remove only that frame's history entries and
skip both PacketBuffer and RtpFrameReferenceFinder cleanup.

Track the newest-media watermark on the H.264/H.265 receive path as
well, so the same validation applies on every codec path.

Add focused RtpVideoStreamReceiver2 coverage and an end-to-end VP9
flexible-mode regression test.

R=asapersson@webrtc.org

Bug: webrtc:556950523
Change-Id: I673ccb0397cb9df891cc4882503c0df5c926d632
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/499540
Commit-Queue: Danil Chapovalov <danilchap@webrtc.org>
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Reviewed-by: Åsa Persson <asapersson@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#48529}