What

Add tests/support/mock_quic_peer_loop.{h,cpp} plus one demo TEST_F. The peer receives one client QUIC Initial datagram, derives QUIC v1 initial keys from the client's original DCID (RFC 9001 Section 5.2), and replies with a header-protected server Initial whose payload contains a stub crypto_frame (type 0x06). This makes quic_socket::process_crypto_frame reachable from a hermetic test for the first time.

Change Type

• Test infrastructure (no src/ behavioral changes)

Why

Part of #1074 (Phase 2C). Continues the work of #1075 (Phase 2A, mock_h2_server_peer) and #1102 (Phase 2B, mock_grpc_server_peer).

Without a peer that can pass the initial-keys gate at quic_socket.cpp:527, the existing mock_udp_peer::make_quic_initial_packet_stub cannot drive any frame-dispatch branch in process_frame -- the entire crypto_frame / stream_frame / ack_frame switch sits behind that gate. This is the structural blocker for the >=80% line / >=70% branch coverage targets in #1062 and #1065.

Related Issues

• Part of infra(test): protocol-aware loopback peers + error injection (Phase 2 of #1060) #1074 (Phase 2 of infra: hermetic transport fixture for HTTP/2 / QUIC / WebSocket / gRPC unit tests #1060: protocol-aware loopback peers + error injection)
• Builds on infra(test): add mock_h2_server_peer for HTTP/2 loopback (Phase 2A of #1074) #1075 (Phase 2A), feat(tests/support): add HEADERS+DATA reply for h2 mock peer (Phase 2A.2) #1082 (Phase 2A.2), infra(test): add mock_grpc_server_peer for gRPC unary loopback (Phase 2B of #1074) #1102 (Phase 2B)
• Unblocks test(http2): expand http2_client.cpp coverage to 80% line / 70% branch #1062 (http2_client.cpp post-connect coverage) and test(quic): expand internal/quic_socket.cpp coverage to 80% line / 70% branch #1065 (quic_socket.cpp post-keys coverage)

Where

No src/ changes. No public-API additions. Phase 2D (friend-test injection for quic_server::handle_packet / websocket_server::handle_new_connection) and Phase 2E (frame_injector) remain open under #1074.

How

Implementation Highlights

The peer mirrors the RAII shape of mock_h2_server_peer and mock_grpc_server_peer:

• Shared io_context from hermetic_transport_fixture
• Dedicated worker thread launched in constructor, joined in destructor
• std::atomic<bool> state flags polled by tests via wait_for(...)
• All four special members deleted
• Hermetic: bound to 127.0.0.1:0 (kernel-assigned port), no DNS, no on-disk secrets

Crypto path reuses production helpers -- no HKDF/AEAD reimplementation:

• internal::quic::initial_keys::derive(dcid) for v1 initial-secret derivation
• internal::quic::packet_builder::build_initial() for long-header construction
• internal::quic::packet_protection::protect() and protect_header() for AEAD + header protection

The reply payload is PADDING + crypto_frame(0x06) carrying 8 zero bytes. process_crypto_frame fires regardless of TLS parse outcome, which is the goal -- driving the branch, not completing the handshake.

One implementation note: packet_builder::build_initial() does not embed the payload-length varint, so the peer code computes ciphertext+tag size and inserts the varint manually before AEAD encryption. Reviewers may want to focus on this section in mock_quic_peer_loop.cpp.

Testing Done

• New demo TEST_F: QuicSocketHermeticTransportTest::ProcessCryptoFrameReachableViaMockQuicPeerLoop
• Asserts peer.initial_sent() and !peer.io_failed() after the client connects. The test's value is that the previously unreachable process_crypto_frame branch executes -- coverage tooling picks that up. The client exposes no observable post-handshake state from outside (the TLS handshake does not complete with a stub crypto frame), so the test asserts peer-side state.
• Local result: ctest -R 'ProcessCryptoFrameReachableViaMockQuicPeerLoop' passes in 9 ms on macOS.

Test Plan for Reviewers

cmake -S . -B build -DBUILD_TESTING=ON -DCMAKE_BUILD_TYPE=Debug
cmake --build build --target network_test_support quic_socket_branch_test -j
ctest --test-dir build -R 'ProcessCryptoFrameReachableViaMockQuicPeerLoop' --output-on-failure

Breaking Changes

None.

Rollback

Revert this PR. No data migration, no public-API impact.

Part of #1074