Asymptotic runtime improvement

[Calvin-L]

This changeset makes asymptotic_runtime a lot more precise without sacrificing much speed. Instead of returning an int, it now returns a DominantTerm that captures the constant factor and exponent of the worst-case runtime.

[Calvin-L]

There are roughly three "prongs" here:

• Using a class with a separate exponent and multiplier makes the result more precise. Previously we were were using 1000 to mean "roughly O(n)", which doesn't quite feel right since any number you choose is going to be dominated by a O(n) term. Now we can differentiate O(1000) and O(n).
• A lot of fishy uses of max were replaced with +. This makes the code easier to read and probably more correct.
• All constant-time terms like O(1000) and O(20) are reduced to O(1) (the if res.exponent == 0 check at the end of asymptotic_cost). This essentially says "I see all constant-time expressions as identical; let other parts of the cost model differentiate these."