What

Summary

Adds branch-coverage focused unit tests for src/protocols/quic/connection.cpp to drive the 62.6% line / 34.4% branch baseline (run 25620254919) toward the >=80% line / >=70% branch acceptance threshold for #1145.

Change Type

• Test addition (no src/ behavioural change)

Affected Components

• src/internal/protocols/quic/connection.h -- adds a forward declaration of quic_connection_test_access and a friend class line, both gated by #if defined(NETWORK_ENABLE_TEST_INJECTION). Production builds with BUILD_TESTS=OFF compile byte-identical.
• tests/support/quic_connection_test_access.h -- new test-only header (header-only, no .cpp) exposing static forwarders to private dispatch helpers and state seeders. Mirrors the pattern of http2_client_test_access.h (infra(test): pivot to friend-injected process_frame for hermetic dispatcher coverage (Round 6, Part of #1106, #953) #1115) and quic_socket_test_access.h (test(quic): expand quic_socket.cpp coverage to 70%/50% #1122).
• tests/support/network_test_friends.h -- forward declaration of the new probe.
• tests/unit/quic_connection_branch_test.cpp -- ~50 new TEST / TEST_F cases.
• tests/CMakeLists.txt -- registers network_quic_connection_branch_test with network::test_support linkage.

Why

Problem Solved

Round 1-6 attempts on #953 produced 0pp delta because tests passed CI without exercising the dispatcher branches: the TLS-1.3 handshake under coverage instrumentation exceeds the wait_for(3s) budget (PR #1111 diagnosis), so every test that depends on connected state silently times out before reaching the branches under measurement.

This PR sidesteps the handshake by directly invoking the private dispatch helpers through the friend-class pattern proven in #1116 and #1122, targeting the largest single-file branch gap in the repository (655 unhit branches in connection.cpp).

Related Issues

• Closes test(quic): expand connection.cpp coverage to 80%/70% (Part of #953) #1145
• Part of Expand unit test coverage from 40% to 80% #953

Who

Reviewers

• @kcenon

Required Approvals

• Owner approval

When

Urgency

• Normal

Target Release

Round 7 coverage uplift for #953.

Where

Files Changed

API / Source Changes

• No public API change.
• The only src/ change is the additive friend class declaration in connection.h under NETWORK_ENABLE_TEST_INJECTION.

How

Implementation Details

The new quic_connection_test_access exposes:

1. Direct dispatch helpers: handle_frame, process_frames, build_packet, generate_ack_frame, handle_loss_detection_result, generate_probe_packets, queue_frames_for_retransmission, update_state, get_pn_space (const + non-const).
2. State seeders: set_state, set_handshake_state, set_close_sent, set_close_received, set_idle_deadline, set_drain_deadline, push_pending_crypto_*, push_pending_frame, seed_sent_packet.
3. Observers: pending_frames_size, pending_crypto_*_size, application_close, close_received, sent_packets_size.

Coverage areas exercised (each TEST targets one or more branches identified by inspecting coverage_html/protocols/quic/connection.cpp.gcov.html):

• handle_frame std::visit arms: padding, ping, ack (seeded sent_packets to drive the erase loop), crypto, max_data, max_stream_data, max_streams (bidi + uni distinct arms), connection_close (transport + application), handshake_done (client promotion + server no-op), reset_stream / stop_sending unknown-stream arms, new_connection_id, retire_connection_id error propagation, data_blocked / stream_data_blocked / streams_blocked / path_challenge / path_response / new_token catch-all arm.
• process_frames empty payload + malformed payload paths.
• build_packet no-keys early-return for all three encryption levels.
• generate_ack_frame nullopt arm and the OR-branch (largest_received == 0 && ack_needed == true).
• get_pn_space switch arms for both overloads including the initial/zero_rtt aliasing.
• update_state no-op when handshake not complete.
• close / close_application re-entry guards (already closing / draining / closed).
• enter_closing / enter_draining no-op arms when already in closing / draining / closed.
• on_timeout drain-timeout, idle-timeout, no-timeout paths; next_timeout closed (nullopt) and draining (drain_deadline) arms.
• handle_loss_detection_result: all three loss_detection_event arms (none, pto_expired with PING frame assertion, packet_lost with retransmission assertion) and both ECN arms (congestion_signal, ecn_failure).
• generate_probe_packets encryption-level cascade (initial / handshake / application) with and without pending crypto data.
• queue_frames_for_retransmission per-variant arms: padding / ack (not requeued), crypto (per level routed to matching queue), stream (skipped -- natural retransmit), ping / new_token / handshake_done (requeued), all six flow-control frames (requeued), reset_stream / stop_sending (requeued), new_connection_id / retire_connection_id (requeued), path_challenge / path_response (skipped), connection_close (skipped).
• start_handshake error guards (server rejection, non-idle rejection); init_server_handshake client rejection.

Testing Done

• Local compile not feasible in the verify-first sandbox (no cmake/gcc available). Local coverage delta will be measured by CI's Coverage Analysis check on this PR and validated by the post-merge develop coverage workflow run per the acceptance criteria in test(quic): expand connection.cpp coverage to 80%/70% (Part of #953) #1145.

Test Plan for Reviewers

1. CI must pass on this PR including the Coverage Analysis check.
2. Post-merge: trigger the develop coverage workflow and verify src/protocols/quic/connection.cpp measures >=80% line AND >=70% branch.

Breaking Changes

• None.

Rollback Plan

1. Revert this PR.
2. No data / migration / configuration impact.

Checklist

• Test additions only; the only src/ change is the additive friend declaration gated by NETWORK_ENABLE_TEST_INJECTION.
• Self-review completed.
• No sensitive data exposed.
• Commit is atomic and well-described.
• Closes #1145 keyword present in PR body.
• PR targets develop.