fix(task): reject boolean max_steps values during task validation

[Adityakk9031]

Summary

Follow-up to #161 as requested by @jeqcho in #161 (comment).

Because bool is a subclass of int in Python (isinstance(True, int) == True), Task(max_steps=True) previously bypassed validation and was silently treated as a 1-step horizon (max_steps=1).

This PR explicitly checks isinstance(self.max_steps, bool) in Task.__post_init__ and raises ConfigError.

---

Changes

• inspect_robots.task.Task: Reject boolean max_steps values during __post_init__ validation.
• tests/test_types_spaces.py: Added parameterized test test_task_rejects_invalid_steps_horizon testing [True, 0, -1].
• CHANGELOG.md: Added entry under ## [Unreleased] -> ### Fixed.

---

Checklist

• pre-commit run --all-files passes
• Tests added/updated
• Coverage stays at 100%
• ruff check . and ruff format --check . pass
• Strict mypy passes
• CHANGELOG.md updated under "Unreleased"

---

Related

Follow-up to #161 (comment) / #160

[Adityakk9031]

@jeqcho follow up pr

[jeqcho]

Thanks for the quick follow-up on the #161 review note — this closes the bool-as-int loophole cleanly, and the scope is exactly what was asked for.

What I checked:

• Correctness (src/inspect_robots/task.py:78): isinstance(self.max_steps, bool) or self.max_steps < 1 is right. True (previously accepted as a silent 1-step horizon) is now rejected, and False — which the old < 1 check already caught — still raises the same error. Putting the isinstance check first also mirrors the structure of the max_seconds guard directly below it (task.py:80-84), so the two horizon validations now read the same way.
• Error message: reusing the existing "max_steps must be >= 1, got True" message rather than inventing a new one keeps things consistent and avoids churn in anything matching on it. (Strictly speaking True >= 1 is numerically true, but the sibling max_seconds message has the same quirk for True, so this is the established convention here.)
• Tests (tests/test_types_spaces.py:147-158): the parameterization [True, 0, -1] covers the new bool branch plus both sides of the numeric boundary, and the test mirrors the style of test_task_rejects_invalid_seconds_horizon right next to it (local imports, pytest.raises(..., match=...)). Nice symmetry.
• CHANGELOG: entry is accurate and correctly placed under ## [Unreleased] / ### Fixed.
• Verification: full CI matrix is green; tests in tests/test_types_spaces.py also pass locally (30 passed).

Two optional nits, neither blocking:

1. src/inspect_robots/task.py:78 sits at 99 characters against the 100-char limit. Wrapping the condition across lines like the max_seconds check below it would make the two guards visually parallel, but ruff is satisfied as-is.
2. Not for this PR, but the same bool-is-int hole still exists in Epochs.__post_init__ (task.py:31-33): Task(epochs=True) currently validates as a 1-epoch spec. Might be worth a tiny follow-up issue for the same strictness there.

Approving — thanks for the contribution!