COMP: Install the tools the analysis jobs actually invoke

[hjmjohnson]

Two of the five Build and Test jobs have never run to completion. Neither failure is in the code under review — both die installing dependencies.

Job	Failure	Cause
sanitize-clang-linux	exit 127, scan-build: command not found	clang-tools never installed
rel-clang-macos	exit 1, No available formula with the name "sed"	sed is not a Homebrew formula

-  sudo apt-get install -y cmake valgrind help2man
+  sudo apt-get install -y cmake valgrind help2man clang-tools

-  brew install cmake sed help2man
+  brew install cmake gnu-sed help2man

Why this went unnoticed

Both defects date from the 2025 conversion to GitHub Actions and have never regressed, because the jobs never ran. The workflow's trigger named main while this repository's branch is master, so Build and Test did not execute on master or on pull requests until #77 corrected it. The first real runs then exposed the dependency lists as they had always been.

macOS ships BSD sed; the GNU one the dashboard scripts expect is gnu-sed. scan-build ships in clang-tools on Debian and Ubuntu, not in clang itself.

What is still red after this, and why

Job	State	Blocker
use_prefix-gcc-linux	green	—
sanitize-clang-linux	fixed here	—
rel-clang-macos	fixed here	—
coverage-gcc-linux	red	three -Wsign-compare warnings, all fixed by #51
valgrind-gcc-linux	red	unclear; see below

coverage-gcc-linux reports 345/345 tests passing, 0 compiler errors and 2 compiler warnings, then fails the check at cmake/nifti_common.cmake:472, which treats any warning as fatal. Reproduced with that job's own flags on Ubuntu 24.04 / GCC 13.3, the warnings are:

nifti2/nifti_tool.c:888    comparison of integer expressions of different signedness [-Wsign-compare]
nifti2/nifti_tool.c:2370   comparison of integer expressions of different signedness [-Wsign-compare]
niftilib/nifti1_tool.c:726 comparison of integer expressions of different signedness [-Wsign-compare]

#51 changes exactly those three lines. Merging it turns this job green without relaxing the gate.

valgrind-gcc-linux is the one that needs investigation rather than a fix. It reports 331/331 tests passing, zero compiler warnings and zero errors, and still fails at nifti_common.cmake:472. None of the five conditions in that if obviously matches, so which one fires should be established before anything there is changed.

Note on the leak fixes that just merged

#62, #60 and #69 do not change the memcheck result. Measured on Ubuntu 24.04 / GCC 13.3, Debug, cifti and fsl on, master and master plus all three both report Potential Memory Leak - 177 across the same thirteen tests with identical per-test counts.

Those changes are still correct. They free memory on error paths, and the suite does not take error paths — consistent with there being no malformed-input corpus. Whatever valgrind-gcc-linux is reporting, it is not what those three PRs addressed.

[hjmjohnson]

The commit messages in this range were rewritten to remove trailers that do not belong in permanent history: Co-Authored-By: naming an AI tool, and Claude-Session: URLs that resolve for nobody. Only messages changed — the tree at the tip of master is byte-identical, and author, committer, and dates are preserved.

This PR's commit on the rewritten master:

• 31474388dc COMP: Install the tools the analysis jobs actually invoke

The SHA recorded above by GitHub is from the pre-rewrite history and no longer resolves.