[BUG] (Barretebberg) batch_mul in cycle_group does not handle 'mixed' constant-ness

[guipublic]

Problem Statement

Doing a batch_mul() with a point whose abscissa is a witness and the ordinate is a constant value is crashing because the point, being not constant (due to x), is assumed to have witnesses for both abscissa and ordinate.
This can be seen in _unconditional_add_or_subtract():
Constantness is true only if all are constants:
is_constant() const { return x.is_constant() && y.is_constant() && _is_infinity.is_constant(); }
But it if is false, the code uses both witness index:

  // Both points are witnesses - create result witness and construct ECC add constraint
        result = cycle_group(witness_t(context, x3), witness_t(context, y3), /*is_infinity=*/false);

        context->create_ecc_add_gate(bb::ecc_add_gate_<bb::fr>{
            .x1 = x.get_witness_index(),
            .y1 = y.get_witness_index(),
            .x2 = other.x.get_witness_index(),
            .y2 = other.y.get_witness_index(),
            .x3 = result.x.get_witness_index(),
            .y3 = result.y.get_witness_index(),
            .sign_coefficient = is_addition ? 1 : -1,
        });

Proposed Solution

This case should only be taken if ! x.is_constant() && !y.is_constant(), and the mixed cases need to be taken into account.

Example Use Case

Run the TestMSM unit test from #16663, but change this value:
msm_constrain.points[0].value = fr(0); // bad input
(right after having defining msm_constrain.points)
This will cause the 'valid_point' (computed inside create_multi_scalar_mul_constraint()) to have x as a witness and as a constant.

Alternative Solutions

No response

Additional Context

No response