tests: close small real coverage gaps in outbox/logger; document 3 dead default: arms

[Yaraslaut]

Investigated 8 small (1-3 gap-unit) flagged files. 2 real gaps closed
with tests, 3 unreachable default: arms documented, 3 already fully
resolved (no action needed -- confirmed already covered/documented by
prior rounds). Also fixes a real root-cause coverage-tooling bug
found while doing this (see below).

Closed with real tests

• outbox.hpp: OutboxRelay::relay()'s null-sink warning path --
  every existing test paired a null sink with a non-empty
  drainOutbox(), which would dereference the null sink (real UB, not
  exercisable). Pairs it with an empty drainOutbox() instead, keeping
  relay() on its early-return path (after the warning fires, before
  any sink use). The sink dereference itself remains tracked by
  LASTRADA-Software/morph#95, unchanged.
• logger.hpp: logFormat's (the std::format_string template
  overload, e.g. logWarn(fmt, args...)) suppressed-below-threshold
  branch -- every existing test only called the plain
  std::string_view overloads.

Documented as unreachable (no test forced)

Three closed-enum-to-string helpers each have a default: arm that
only exists to satisfy the compiler's exhaustiveness check for an
out-of-range static_cast -- every real enumerator is already handled
explicitly, mirroring forms.hpp's already-established
ruleKindName pattern:

• i18n.hpp's fieldSlotName (FieldSlot)
• layout.hpp's groupKindName (GroupKind)
• reconnect_coordinator.hpp's reconnectOutcomeName (ReconnectOutcome)

Already resolved, no action needed

• network_monitor.hpp -- confirmed already covered/documented by the
  prior round (deadlock-on-destructor case).
• executor.hpp -- both flagged call sites already exercised
  extensively, including exception-throwing tasks.
• wire.hpp -- already documented as unreachable, tracked by
  LASTRADA-Software/morph#96.

Also: fixes a real coverage-tooling root cause

examples/common/testkit/ mixes real, reusable test-support code
(backend_rig.hpp, db_fixture.hpp, strand_interleaver.hpp, ...)
with actual Catch2 test files (test_event_poller.cpp,
test_presenter.cpp, test_fault_proxy.cpp, ...) in the same
directory. scripts/coverage.sh's SOURCES+=(examples/common) swept
both kinds in indiscriminately -- a TEST_CASE body's own untaken
assertion/lambda branches (a REQUIRE's fail arm, a "must not run"
callback proving itself unreachable) were being measured as if they
were product code.

This is the exact "phantom uncovered branch" pattern this session
repeatedly had to hand-verify file by file across several PRs before
concluding each one was tooling noise, not a real gap. Adds
-ignore-filename-regex='.*/testkit/test_[^/]+\.cpp$' to all four
llvm-cov invocations, verified directly against the real llvm-cov
binary/version this repo's CI uses: correctly excludes every
test_*.cpp under testkit/ while leaving every real testkit helper
measured exactly as before. scripts/aggregate_lcov_branches.py's own
output is confirmed byte-identical before/after (this change doesn't
touch that script).

Also investigated LASTRADA-Software/morph#92 (proposed DA:
line-record aggregation, mirroring the existing BRDA: branch
aggregation) while here -- checked its premise against real downloaded
CI data and found llvm-cov's lcov exporter already merges line-hit
counts across template instantiations correctly (no duplicate DA:
records found for any of the file/line examples the issue names).
Posted findings there; left the issue open for a second opinion rather
than closing it myself.

Verification

Full morph_tests suite: 1046 test cases, 10038 assertions, all
passing.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!