fix scalar multiply

[Debraheem]

This PR fixes a typo in the accurate auto-diff multiply code.

When MESA did:

accurate_ad_value * scalar

the code accidentally multiplied by the accurate auto-diff value again, instead
of multiplying by the scalar.

So it was effectively doing:

value * value

when it should have done:

value * scalar

The fix changes the multiply routine, so it uses the scalar. This makes
accurate_ad * scalar behave the same way as scalar * accurate_ad.

This could have affected the hydro routines that use accurate auto-diff sums,
like get1_energy_eqn, get1_momentum_eqn, do1_dudt_eqn, and
do1_turbulent_energy_eqn. But those routines avoided this bug because they
either convert back to normal auto-diff before scaling, or use
the safe multiply direction. That small syntax difference prevented this bug from
breaking anything, because scalar * accurate_ad_value worked, while
accurate_ad_value * scalar was the broken operator overload.

[warrickball]

Yikes, good catch! I'm running the test suite on bluebear now.

Do we not have unit tests to catch this? Presumably this error would jump out for appropriate choices of op1 and op2.

I don't think I was aware of this type and am also wondering why it's in num instead of auto_diff.

[VincentVanlaer]

I would say num is a fine location for this, since it is also a specific algorithm.

[Debraheem]

I think this would have come out if the operators were switched, but i don't think it affects anything at the moment.