ppvm-sym: Term arithmetic bugs (half, AddAssign<f64>, MulAssign)

[david-pl]

Found while looking at Term as a Coefficient for #219. All four reproduce on main (da073eb); tests below fail today.

1. Term::half() ignores self

coeff.rs#L9 returns Term::from_f64(0.5) instead of self / 2. PauliSum projections call v.half() (proj.rs#L16, #L35), so symbolic measurement projections replace the coefficient with the constant 0.5.

Fix: self.clone() * 0.5.

2. AddAssign<f64>: One + const is a no-op

add.rs#L24: the Inner::One arm builds a Sum but never assigns it to self.inner, so sin(x) += 1.0 leaves sin(x) unchanged.

3. MulAssign<Term>: One * Sum is a no-op

mul.rs#L172: new_sum.mul_term(...) result is dropped, self keeps its One value.

4. MulAssign<Term>: Sum * Sum drops negative constants

mul.rs#L114 and #L121 guard with c0 > min_eps instead of c0.abs() > min_eps, so cross terms with a negative constant vanish, e.g. (sin x − 1)(cos x − 1). The guard can simply be removed: Sum::add_term already drops |coeff| < min_eps.

5. (not verified by a test) MulAssign<Prod> drops phase

mul.rs#L64 merges sin/cos powers but never combines phase, which would affect the ComplexCoefficient path.

Reproducing tests

use ppvm_sym::Term;
const X: [f64; 1] = [0.3];
fn v(t: &Term) -> f64 { t.eval(&X).unwrap() }

#[test]
fn half_ignores_self() {
    use ppvm_traits::traits::Coefficient;
    assert!((v(&Term::from_f64(4.0).half()) - 2.0).abs() < 1e-12);
}

#[test]
fn one_plus_const() {
    let mut t = Term::var(0).sin();
    t += 1.0;
    assert!((v(&t) - (X[0].sin() + 1.0)).abs() < 1e-12);
}

#[test]
fn one_times_sum() {
    let mut s = Term::var(0).sin();
    s += Term::var(0).cos();
    let mut t = Term::var(0).sin();
    t *= s;
    assert!((v(&t) - X[0].sin() * (X[0].sin() + X[0].cos())).abs() < 1e-12);
}

#[test]
fn sum_times_sum_negative_const() {
    let mut a = Term::var(0).sin();
    a += Term::from_f64(-1.0);
    let mut b = Term::var(0).cos();
    b += Term::from_f64(-1.0);
    assert!((v(&(a * b)) - (X[0].sin() - 1.0) * (X[0].cos() - 1.0)).abs() < 1e-12);
}