scripts: measure_hero_drift.py exits 0 on size mismatch or unknown --only, and crashes on bad input

[TMHSDigital]

Why it matters

scripts/measure_hero_drift.py is how the repo proves that committed hero stills match what the code renders now (#200). Several of its paths report success when nothing was measured, which is the kind of assertion that "witnesses nothing" by the repo's own falsification standard.

Evidence

• scripts/measure_hero_drift.py:163-165: when the rendered and committed images differ in size, the row lacks mean_abs. The loop prints a message and continues. The row counts as neither drifted nor failed, so the script exits 0. A size mismatch is itself drift.
• An unknown --only NAME matches nothing, measures zero heroes, and exits 0.
• :131: …stdout.splitlines()[0] raises IndexError if blender --version fails or prints nothing, which surfaces as a traceback instead of a clear error.
• :78-79: args.index("--output") / args[i + 1] raises IndexError when a README command ends in a bare --output.

Suggested approach

• Treat a size mismatch as a failure and include it in the non-zero exit.
• Exit non-zero with a clear message when --only names match no example or showcase piece, or when zero heroes were measured.
• Guard the version probe and argument parsing with explicit error messages and a documented exit code.
• Add unit tests for these paths. Today tests/smoke/test_harness.py is the only unit test, and the scripts/ tooling has none.

Done when

• Each of the four cases above exits non-zero with a readable message.
• Each case is covered by a test that was seen failing against the current code.