Before reading your comment, I probably would have agreed with you that isOdd was overkill.

After reading your comment, I'm starting to think that isOdd is not only more readable but safer. I don't think I ever thought about n % 2 not handling negative numbers, and I'm not sure I would immediately recognize n & 1 as equivalent to isOdd. In languages without truthiness, n % 2 == 0 is longer than isOdd(n). Likewise, even with truthiness, n % 2 == 1 is longer than isEven(n).

I suspect that the real problem is the cost of adding dependencies, in terms of package size, performance, and security. All these problems could be solved using linking, inlining, and auditing.