test, raise library line coverage above 95%

[steven-varga]

Goal

Raise h5cpp library line coverage from ~82% to >95%.

Context — measure the right denominator first

The CDash Coverage step was counting vendored thirdparty, the test harness, and example TUs (9276 LOC counted vs 2453 real library lines), burying the true figure. Fixed under #278 via CTEST_CUSTOM_COVERAGE_EXCLUDE in CTestCustom.cmake, mirroring the lcov --remove list already used for Codecov. With that scope, the real numbers are:

• Lines: ~82% (2019/2453)
• Functions: 95.2% (1387/1457)

Measured locally: gcc-14 --coverage, Debug -O0, HDF5 1.12.3, C++20, 54/54 tests pass.

Gap is concentrated (434 uncovered lines)

Header	Uncovered	Rate	Share
H5Zpipeline_pool.hpp	184	0.0%	42%
H5cout.hpp	~87	37.4%	20%
H5Dappend.hpp	~31	85.6%	7%
H5Zpipeline.hpp	~29	79.1%	7%
H5Dread.hpp	~26	78.7%	6%
H5Ialgorithm.hpp	~16	55.6%	4%
H5Dwrite / H5capi / H5Dopen	~33	—	8%

H5Zpipeline_pool.hpp (the threaded filter pool) is the keystone: 0% and 42% of the entire gap.

Plan

1. Dedicated test for H5Zpipeline_pool (threaded filter pool) — model on H5Zpipeline_parallel_read.cpp.
2. H5cout print/dump-path test.
3. H5Zpipeline edge branches.
4. H5Dread / H5Dappend edge paths.
5. Re-measure; top up H5Ialgorithm / H5Dopen as needed to clear 95%.

Acceptance

• Library line coverage >95% on the CI coverage job (post-ci, off-CI CDash submissions from any architecture #278 scope).
• All tests pass; no thirdparty/test/example lines in the denominator.

[steven-varga]

Status — reachable coverage raised 82.3% → 94.0%; >95% gated on #286

Done (branch 285-raise-coverage-above-95, 56/56 tests pass):

• New/extended tests: H5cout handle printers (37→94%), dfs/bfs traversal (55→75%), rank-2 chunked pipeline write path, the synchronous high_throughput write/open/read dispatch (H5Dread 79→90%, H5Dopen 60→80%), and attribute/create/fletcher32 edges.
• Excluded the unreachable H5Zpipeline_pool.hpp (180 dead lines, see fix, carry per-file worker pool in a fileno-keyed singleton registry (H5Fget_access_plist strips FAPL props) #286) from the Codecov pipeline; CDash-side exclusion is in ci, off-CI CDash submissions from any architecture #278.

Reachable line coverage: 94.0% (2270/2416), functions 96.3%.

Why not 95% yet — same root cause as #286

The remaining gap is dominated by the #286 dead worker-pool branches that live inside H5Dwrite/H5Dread/H5Dappend (~44 lines). Coverage excludes are file-granular, so these can't be dropped without dropping the whole (tested) files. The clean way past 95% is fixing #286 — that makes ~224 lines live (already exercised by the #263 + new high_throughput tests) and clears 95% over the full denominator with zero exclusions.

Accepting 94% as the landed result here; #286 is the follow-up that completes the >95% goal.