tests: split elf coverage into no-reloc control and reloc gate

[jlinenkohl]

Summary

• Split ELF tests into:
  • no-reloc control path ([ELF][no-reloc])
  • relocation capability gate ([ELF][reloc])

Why

• Separates stable baseline behavior from dynamic relocation capability checks.

Validation

• cd tests && bash run_unit_tests.sh -R test_elf

Depends on

• linux: fix readlinkat guest copy length handling #67, elf: apply R_X86_64_RELATIVE addend semantics #68, elf: add .relr.dyn bootstrap relocation support #69 (for green relocation gate behavior)

Stack Context

• Final capstone test-gate PR for the relocation stack.

Test Evidence

• Date: 2026-04-02
• Branch-level validation source: phase14_audit baseline sweep
• Full unit harness: cd tests && bash run_unit_tests.sh -> 8/8 passed
• Integration tinytest lane: (cd guest/tests && bash build.sh) && ./build/tinytest guest/tests/glibc_test -> passed

PR-Scoped Command

• cd tests && bash run_unit_tests.sh -R test_elf
• Stacked expectation: final capstone, expected fully green after linux: fix readlinkat guest copy length handling #67/elf: apply R_X86_64_RELATIVE addend semantics #68/elf: add .relr.dyn bootstrap relocation support #69

[fwsGonzo]

I don't quite understand this PR. Most unit tests are already statically linked binaries, so adding a return 123; test is not adding anything new or testing any new code paths.
The helper for current working directory is only useful to avoid a leak using https://linux.die.net/man/3/get_current_dir_name which needs to freed. However, it's used only once, and it's not a concerning issue. Unless the test-suite is run with lsan, I guess. I'd merge just the cwd change.

[fwsGonzo]

Adding a new test that just returns 123 is not a new codepath. Most unit tests are static ELFs already. We could maybe merge the CWD change to avoid a leak with get_current_dir_name().

[jlinenkohl]

I've pulled this back to draft for now as this was submitted too soon as a batch with the others. I had added some test gates for working on relocation support and this was not ready to go upstream. This was pushed up because of the CWD fix but was not supposed to include the rest.

I do appreciate your feedback though and have noted to include that minor fix in a subsequent update.

Also, about to propose a merge of a different fix in PR71 that caused a SIGSEGV. I discovered it while doing relocation work.