Optimize SparseTerm comparisons and constant checks

[ANtutov]

This change reduces redundant work in SparseTerm hot paths without altering semantics. In partial_cmp, computing degree() for both sides once and reusing the results avoids repeated summations during term sorting invoked by SparsePolynomial::from_coefficients_vec, which can be significant for large term lists. In is_constant, an early return when the term is empty eliminates unnecessary degree aggregation for the common constant case, while retaining a fallback degree() == 0 check to remain correct for terms constructed without normalization, such as SparseTerm(vec![(i, 0)]), which are present in tests.