[hugojosealmeida]

Summary

Free file metadata discarded by successful lsof_gather calls while preserving transferred result names.

Changes

• Release untransferred names and filesystem strings.
• Make the Solaris fsdev fallback an owned copy.
• Add regression coverage for selected and filtered files.

Verification

• make check TESTS=tests/LTfilemeta: pass.
• Test and lib/lsof.c instrumented with ASan + UBSan: pass.
• Darwin leaks: 84 leaks / 7,632 bytes before; 3 / 4,128 after. Metadata allocation stacks disappeared; separate Darwin gather-buffer leaks remain.

---

This PR is AI Assisted.

Closes #379

[hugojosealmeida]

I have reproduced on Darwin: three filtered gathers reported 84 leaks / 7,632 bytes. With the fix, that dropped to 3 leaks / 4,128 bytes; the allocation stacks for nm, fsdir, and fsdev disappeared.
The remaining leaks seems to be Darwin gather buffers outside this fix. I have added a test that also confirms that selected names survive context destruction and the filtered FD is not returned.

[hugojosealmeida]

The new test requires an exact pathname match and so maybe the failure is likely a portability issue in the test I added: the existing LTbasic2 explicitly calls stat() after file creation to make FreeBSD expose the pathname. I will investigate.

[hugojosealmeida]

Found it, the test assumes name is a raw pathname, but FreeBSD appends filesystem information. Its exact strcmp()therefore rejects a valid result. The fix on the test changes that comparison tostrstr()` (and adds a comment).

[jiegec]

Please rebase, and I think it is unnecessary to write a test for this.