deps: bump moq-noq for the lost handshake ACK fix

[kixelated]

Quest: quest/m1/test-flakes-2/handshake-ack.md (deleted here).

Draft until moq-dev/noq releases 2.0.5. The fix is in moq-dev/noq#38. This PR still needs the pin bump, below.

Problem

MOQ_SHAPER_SEED=13876846873975217772 cargo test -p moq-relay --test drills bursts_cross_a_cluster::impaired fails every run on main with publisher connect timed out (2 of 2 today). The client's ACKs of the server's first Initial are all lost, and noq writes an ACK only while PendingAcks::can_send() holds, so the client's probes in between carry none. The server keeps resending its two-datagram flight, the shaper drops the tail each time, and the 10s connect budget runs out. The quest has the full trace.

Fix

moq-dev/noq#38 makes every Initial and Handshake packet repeat the pending ACK ranges, the handshake half of what quinn-rs/quinn#2747 does (and moq-quic already has). With it patched in locally:

• The failing seed passes 3 of 3 runs, 6.3 to 6.8s.
• 50 random seeds pass, 50 of 50.

This PR adds the same mock-time test to moq-quic, lost_handshake_ack_repeats_in_next_packet, so the switch off noq keeps the behavior. It passes on moq-quic as is and fails if can_send_with_other_frames returns false.

The PTO cascade the quest asked about (an ACK resets pto_count while the Handshake PTO deadline is already past, and noq refires it in one handle_timeout call, 6 times and 12 probes in the trace) is left alone. moq-quic's handle_timeout fires each timer at most once per call, so it does not cascade, and the switch removes it.

Remaining

Once moq-noq-proto 2.0.5 is on crates.io:

1. Cargo.toml: moq-noq-proto and web-transport-moq to 2.0.5.
2. cargo update -p moq-noq-proto -p moq-noq -p moq-noq-udp -p web-transport-moq.
3. Rerun the seed above, then mark ready.

API: none. Wire: none (noq may now include an ACK frame in an Initial or Handshake packet that previously omitted it).

Tests

• cargo test -p moq-quic: 447 + 3 passed. cargo clippy -p moq-quic --tests: clean.
• The drill runs above, with moq-noq-proto patched to the noq#38 branch.

(Written by Claude Opus 5.5)

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[kixelated]

Outcome: the fix is moq-dev/noq#38 (draft), which repeats pending Initial and Handshake ACK ranges in every packet. Patched in, the failing seed passes 3 of 3 and 50 of 50 random seeds pass. This PR carries the matching moq-quic regression test and the quest deletion; it needs the 2.0.5 pin bump once noq is released.

Open decisions for the maintainer:

1. Backport Relay support #38 to noq 1.3.x? Recommend no: Implement (forked) moq-transport-00 #34 was not backported either, and this repo pins 2.0.x.
2. The PTO cascade was left unfixed in noq, reading the quest's "fix it only if moq-quic does the same" as "only if moq-quic also cascades". It does not, since quinn's handle_timeout fires each timer once per call. If the intent was instead "make noq match moq-quic", it is a small change in noq's handle_timeout loop. Recommend leaving it to the switch.

(Written by Claude Opus 5.5)