fix(rtp): reset PaddingSize when Padding bit is unset (closes #140)

[Booyaka101]

Summary

rtpUnmarshalPayload reuses the same *rtp.Packet across reads. When a packet with Padding=true is followed by one with Padding=false, the second packet's PaddingSize field is left stale at the first packet's value. The downstream payload-size calculation (rtp_packet_reader.go:153, rtp_session.go:265) subtracts pkt.PaddingSize unconditionally, so the stale value truncates the second packet's payload.

Reported and root-caused by @brandon-hc in #140.

Fix

Reset p.PaddingSize = 0 in the else branch of rtpUnmarshalPayload, matching pion/rtp's own Unmarshal behavior at packet.go:240-243.

 	end := len(buf)
 	if p.Header.Padding {
 		p.PaddingSize = buf[end-1]
 		end -= int(p.PaddingSize)
+	} else {
+		p.PaddingSize = 0
 	}

The downstream consumers at rtp_packet_reader.go:153 and rtp_session.go:265 are unchanged — the fix is at the unmarshal layer where the field gets populated.

Regression test

TestRTPUnmarshalResetsPaddingSize exercises the exact reuse scenario: marshal a padded packet, unmarshal into pkt, then marshal an unpadded packet and unmarshal into the same pkt. Asserts pkt.PaddingSize == 0 and the full payload survives.

Verification

State	Platform	Result
Pre-fix	Windows + Go 1.26	FAIL: expected 0x0, actual 0x4 at rtp_parse_test.go:56 — confirms the test exercises the bug
Post-fix	Windows + Go 1.26	PASS (regression test + full go test ./media/ suite, 1.7s)
Post-fix	Linux (golang:1.26 container)	PASS

Out of scope

• The pion pkt.PaddingSize field is now deprecated in favor of pkt.Header.PaddingSize. Diago still uses the legacy field; migrating is a larger structural change and a separate concern.
• Video support / refactoring discussed in the issue thread is orthogonal — this fix is correctness for any RTP usage, not video-specific.

[emiago]

I am having hard time to follow on this, but we can merge this as I see.

[Booyaka101]

Sure — boiling it down:

Diago reuses one *rtp.Packet across reads. rtpUnmarshalPayload only writes p.PaddingSize on the padded branch, so when a padded packet is followed by an unpadded one, PaddingSize stays stuck at the previous packet's value. Downstream code (rtp_packet_reader.go:153, rtp_session.go:265) subtracts pkt.PaddingSize unconditionally, so that stale value chops bytes off the new packet's payload.

The one-line fix sets p.PaddingSize = 0 in the else branch — pion/rtp's own Unmarshal does the same thing at packet.go:240-243, which is why the field never goes stale on their side. The regression test marshals a padded packet then an unpadded one into the same pkt and asserts the payload survives; it fails on main and passes with the fix.