Clarify/modify bytecode cost function

[kazimuth]

Currently our bytecode "cost function" - what we use to count user bytecode - doesn't account for the fact that some bytecode are more expensive than others on the actual machine underneath the VM. Also, it's all scattered around this file, which will make it hard to implement in various editors.

Also, we may want to adjust and document the method costs of various standard library methods. Note that Math.sin and other trig functions currently cost 50 bytecode, which is pretty cruel in continuous space.

We should have a simple definition that will be clear and easy to implement. We should probably put it in the spec.

See also: bytecode instruction listings.

cc @jbloxham