shell_exec_run: judge the witness argv as a whole value (optional-equality rule)

[gunbai-bot]

One file, one function. The composed floor (merge_group run 37766479936) refused 3 diagnostics from the optional-equality rule at src/v2/test/claim/execution/emit_host_shell_exec_run_equals_eval_test.dag:44-45: the eval leg compared List.first() against string literals, and first() is Optional-typed, so first() == "bash" is refused.

Per the #13581 finding, the tempting Present-pattern rewrite executes FALSE at runtime (first() yields the bare head), so the fix is whole-value equality: the witness argv is judged against its expected list ["bash", "-s"] directly -- no Optional constructed, no first() call, and the condition is exactly the old conjunction (length 2 AND first two elements).

Verified by execution on the runner: claim_batch on the three claims returns PASS on all (eval octet = 1, the equals-eval leg holds, the wrong-oracle leg still refuses), BATCH_EXIT=0 -- not just compile-clean.

[briansrls]

Approved at abfb678. The one-line rewrite is semantically equivalent to the old predicate: equality with ["bash", "-s"] requires exactly two elements in exactly that order, while evaluating the argv producer once and avoiding both Optional-vs-required comparisons and the runtime-false == Present { ... } form. The three claim roles remain intact: the eval oracle yields 1 only for the declared argv, the native/eval agreement consumes that octet, and the complemented oracle must still refuse. The diff is confined to this helper. Exact-head witnesses run 37796383849 is queued; queue only after the aggregate turns green.

[gunbai-bot]

Closing as superseded: main fixed the same witness independently in d0f2067 (match the Optional from first(), the runtime-sound form per #13581). This PR now conflicts and adds nothing. — sent from lively-ram-153