BUG: Do not write to the stream that failed to open

[gdevenyi]

open_write_stream() returns NULL when its fopen() fails, and reports it. Both callers then used the result without testing it.

$ cifti_tool -input c.nii -disp_cext -output /nonexistent-dir/x.txt
** failed to open '/nonexistent-dir/x.txt' for writing
Segmentation fault (core dumped)

disp_cifti_extension() passes it to fprintf(). eval_cifti_extension() hands it to axml_set_wstream(), so the NULL becomes the XML library's write stream and every later write follows it. Both now return 1 — the message has already been printed.

Split out of #56, which is about a different defect in one of the same functions (an extension search that never advanced). #56 touches adjacent lines, so if both are in flight, merge #56 first and this rebases.

Note cifti/ is behind USE_CIFTI_CODE, which defaults to OFF, so the standard workflows do not compile this file. Verified locally with -DUSE_CIFTI_CODE=ON.

🤖 Generated with Claude Code

[hjmjohnson]

Rebased onto current master (b4876bf) and added a red-green regression test. Full suite: 363/363 pass.

Worth your call: process() in cifti_tool.c discards the return value of both disp_cifti_extension() and eval_cifti_extension(), so even with this fix the tool exits 0 after failing to open its -output. This PR does not change that, and the test asserts the current exit-0 behavior. If you want the failure propagated to the exit status, that is a one-line follow-up and I can add it here.

Rebase and commit-message cleanup

The ~29 shared-base commits are already on master under new SHAs, so the rebase replayed only the single commit this PR owns. No conflicts.

Message cleanup: removed the Co-Authored-By: trailer naming an AI tool, and rewrapped the body to the project's line limits. Author, committer, dates and tree are unchanged.

Test added, and the red proof

cifti/test_write_failure.cmake, driven by cmake -P rather than a plain add_test, because PASS_REGULAR_EXPRESSION ignores the exit status — a crash that had already printed the diagnostic would pass.

Red, with only the cifti_tool.c fix hunks reverted:

1/1 Test #363: cifti_write_failure ...***Failed
CMake Error at cifti/test_write_failure.cmake:16 (message):
  expected result 0, got 'Segmentation fault'

  ** failed to open '/tmp/prb-u1/cifti/no_such_dir/out.txt' for writing

Green with the fix. The script asserts a normal exit and the presence of the diagnostic, not merely that the command failed. Companion assertion: an openable -output must still be written and must contain MapName, so a fix that refused every write would not pass.

Scope note: the red proof covers the disp_cifti_extension() hunk. The eval_cifti_extension() hunk has no observable difference on this platform — unfixed, axml_set_wstream(NULL) is accepted and the run still exits 0 with the same diagnostic — so it is not separately pinned. It is still the right change.