Fix Quantity * Quantity and / Quantity on compound ratio units - #5478

Quantity.__mul__ delegated quantity*quantity to (quantity*scalar)*unit
and __truediv__ mirrored it. When self's unit is a dimensionless
compound ratio such as nanometer/angstrom, the scalar operation
collapses it to a bare value first, so the later unit operation
reattached only the other quantity's unit. Products dropped a unit
factor (3.0 nm/A squared gave "9.0 nm/A" instead of 900.0) and
quotients inherited the wrong unit entirely (a/a gave 0.1, a/b gave
15.0 instead of 150.0).

Multiply/divide the raw values and combine the units in one Quantity,
then reduce_unit as before. This preserves the existing float
association for ordinary units while fixing the ratio-unit collapse.

Fixes openmm#5445.

Since openmm#5477, scalar * dimensionless-unit expressions evaluate to plain
floats, so 3.0 * (u.nanometer / u.angstrom) no longer produces a
Quantity and the test stopped exercising Quantity arithmetic.  Build
the test values with u.Quantity() directly.

The raw value*value and value/value paths raise TypeError when either
Quantity wraps a plain list or tuple (e.g. Vec3 values). Fall back to
_scale_sequence in that case, matching _change_units_with_factor.

Generated with [Devin](https://devin.ai)