#1109)

* test(support): add post-handshake server-frame hook to mock_h2_server_peer

Adds an optional post_handshake_frames constructor argument that lets
callers supply pre-serialized HTTP/2 frame byte buffers for the mock
peer to write immediately after the SETTINGS-ACK and before the
per-mode request/response handling.

Each entry is written verbatim, bypassing the frame_injector, so tests
can construct precisely-formed PING, GOAWAY, WINDOW_UPDATE, RST_STREAM
or arbitrary unknown-type frames whose wire format frame::parse will
accept and route through the client's process_frame dispatcher.

This unlocks dispatcher-branch coverage paths in http2_client.cpp that
the request-error path (echo_one + injector) cannot reach because
those branches require a server-originated frame on the connected
state, not a corrupted response to a client-initiated request.

Default empty value preserves existing Phase 2A / 2A.2 / 2E.R2
behavior for all current call sites.

* test(http2): guard coverage-flaky tests, add dispatcher branch tests

Phase 2E.R3 follow-up to PR #1108. The Round 2 work landed 6 new TEST_F
cases on the http2_client.cpp branch test, but the post-merge coverage
workflow showed line/branch coverage on http2_client.cpp unchanged at
18.8% / 9.9% (target 60% / 40%).

Root cause: 2 positive-path TEST_F (slow_write handshake, echo_one +
truncate-at-9 response) fail under coverage instrumentation because
lcov/gcov slow the SETTINGS handshake past the 2-3 second wait budget.
The other 4 TEST_F pass but only re-cover early-connect lines that
existing happy-path tests already hit, so they contribute zero net
coverage. handle_ping_frame, handle_goaway_frame,
handle_window_update_frame, handle_rst_stream_frame, and process_frame
default branch remain unreachable.

Two changes:

1. Guard the two coverage-flaky tests with #ifndef NETWORK_COVERAGE_BUILD.
   tests/CMakeLists.txt defines NETWORK_COVERAGE_BUILD on the branch
   test target only when ENABLE_COVERAGE is on, so Debug/Release matrix
   builds keep the slow_write partial-read accumulator and truncate-on-
   connected request-error branches under test, while the coverage
   workflow no longer carries a measured-failing-test signal.

2. Add 7 dispatcher TEST_F cases that use the new
   mock_h2_server_peer post_handshake_frames hook to write
   server-originated frames after the SETTINGS handshake:

   - ServerPingFrameDrivesHandlePingAndKeepsConnectionAlive
   - ServerPingAckFrameIsAbsorbedSilently
   - ServerGoawayFrameFlipsConnectionStateToDisconnected
   - ServerWindowUpdateOnConnectionStreamExpandsWindow
   - ServerWindowUpdateOnUnknownStreamIsSilentlyIgnored
   - ServerRstStreamOnUnknownStreamIsSilentlyIgnored
   - ServerUnknownFrameTypeIsHandledWithoutCrashing

   Each frame is serialized via the production frame class
   (ping_frame, goaway_frame, window_update_frame, rst_stream_frame)
   so frame::parse accepts the wire format and routes the frame to
   the intended handler. The unknown-type case constructs raw bytes
   with type=0xFF to drive either process_frame's default branch or
   the run_io read-error path; both outcomes are RFC-compliant.

Local Debug build: existing pre-merge hermetic instability on macOS
Apple Silicon affects the new tests the same way it affects the
unmodified ConnectCompletesSettingsExchangeWithMockPeer baseline
(documented in PR #1108 body). CI Linux + GitHub macOS runners are
the authoritative verification surface, as before.

Closes #1106

…ixture (#1113)

Scale every hermetic-fixture wait_for budget by NETWORK_COVERAGE_TIMEOUT_MULTIPLIER
so Http2ClientHermeticTransportTest dispatcher cases can complete the SETTINGS
handshake under -fprofile-arcs -ftest-coverage instrumentation. PR #1111
diagnosis confirmed all 7 Round-3 dispatcher TEST_F (and ~30 other fixture-based
tests) failed at wait_for(settings_exchanged(), 3s): 5.1-5.4s observed under
coverage vs 100-300ms in Debug/Release, producing 0pp delta on
src/protocols/http2/http2_client.cpp for three rounds (PR #1108, PR #1109).

Changes:
- tests/support/hermetic_transport_fixture.h: define
  NETWORK_COVERAGE_TIMEOUT_MULTIPLIER (default 1), apply to wait_for default
  timeout argument.
- tests/support/hermetic_transport_fixture.cpp: scale make_loopback_tcp_pair
  accept-handler deadline by the multiplier.
- tests/support/CMakeLists.txt: define
  NETWORK_COVERAGE_TIMEOUT_MULTIPLIER=5 on network_test_support PUBLIC when
  ENABLE_COVERAGE is ON, so all consuming test executables inherit it at the
  point of header inclusion.

No src/ changes. No new TEST_F. Multiplier defaults to 1 outside coverage
builds; Debug/Release matrix observable behaviour is unchanged.

Closes #1112