feat: add exact line shell expectations

[759132989-crypto]

What does this PR do?

Adds expect_exact_lines to kind: shell checks. It compares ordered stdout lines while ignoring trailing whitespace on each line and blank trailing lines. README and CHANGELOG are updated.

Why?

Existing shell expectations could match a flat exact string, regex, or minimum number, but could not express a readable ordered list of exact output lines.

How is it tested?

Added scripts/eval/tests/shell_exact_lines.sh with positive, reordered-line negative, and empty-output cases.

Targeted validation on macOS:

• bash scripts/eval/tests/shell_exact_lines.sh — PASS
• bash scripts/eval/tests/shell_no_expectation.sh — PASS
• bash scripts/eval/tests/shell_safety.sh — PASS
• bash -n scripts/eval/lib/score.sh scripts/eval/tests/shell_exact_lines.sh — PASS
• git diff --check — PASS

Before / after evidence

Before: a YAML list under expect_exact_lines was not recognized as an expectation.

After: the new regression suite prints:

PASS: shell expect_exact_lines compares ordered lines and ignores trailing whitespace/blank lines

---

Checklist

• I ran for t in scripts/eval/tests/*.sh; do bash "$t"; done and all suites passed
• I added a test (or updated one) covering the change
• I updated CHANGELOG.md under ## Unreleased
• I read CONTRIBUTING.md
• (If touching score.sh or attribute.sh) I checked the change works under BSD grep on macOS

The full suite was attempted under Bash 5.2 on macOS. Unrelated environment-dependent suites failed because their opencode stub/path resolution returned exit 127, and one LangGraph test used GNU-only sed syntax. I am not claiming a fully green suite.

Prepared with AI assistance in a human-supervised workspace; the human account owner remains accountable for this submission and review responses.

Closes #31.