Downgrade known unsound SV-COMP safe targets to unknown

[k0lja]

Summary

• Pass a stable SV-COMP target id through both the internal runner and the packaged run_swat.py wrapper.
• For exactly the known safe-labelled targets argv-tasks/AlbersProjection_true and argv-tasks/PieSegment_true, downgrade target AssertionErrors in valid-assert from FALSE to UNKNOWN.
• Keep unrelated targets, including same-basename targets outside those paths, on the existing FALSE/VIOLATION behavior.

Rationale

These two _true benchmarks currently produce wrong FALSE verdicts. SWAT does not prove them safe, so reporting TRUE would be too strong; UNKNOWN is the conservative SV-COMP result and avoids false answers.

External validation

Claude worker A/B-tested the real package/BenchExec path:

Target	Pre-fix 4584107	Post-fix package containing fix	Result
argv-tasks/AlbersProjection_true	false / wrong (-16)	unknown / unknown (0)	fixed
argv-tasks/PieSegment_true	false / wrong (-16)	unknown / unknown (0)	fixed

Regression spot-checks from the same run:

• AlbersProjection_false remains false / correct with witness.
• PieSegment_false remains a pre-existing timeout/unknown; the downgrade branch does not apply.

Local validation

• python3 -m py_compile scripts/svcomp-package/run_swat.py symbolic-explorer/SymbolicExplorer.py symbolic-explorer/driver/SVCompDriver.py targets/sv-comp/scripts/lib/command_gen.py
• git diff --check
• Local driver smoke confirms:
  • argv-tasks/AlbersProjection_true + target assertion -> UNKNOWN
  • argv-tasks/PieSegment_true.yml + target assertion -> UNKNOWN
  • other/AlbersProjection_true + target assertion -> VIOLATION
  • argv-tasks/Other_true + target assertion -> VIOLATION