Add individual function optimization

[Ivorforce]

Possibly builds on top of #76.

It would be nice if functions could be enabled in 3 stages:

• Disabled
• Enabled (saving space)
• Enabled (optimized, sacrificing space)

Currently, optimizing all functions seriously will blow up the binary to over 100mb. But if only 1-5 functions would be optimized, this would result in much more manageable sizes.

This feature is currently not high priority, unless someone needs it.

[Ivorforce]

One way to implement this would probably to have feature flags (like #76) that pertain to whether a function is specifically optimized. It will then use the corresponding functions to make templates expand to either a normal function version or an optimized one.

[Ivorforce]

I don't think this is needed now. in-place correct type math is perfectly optimized now, and if somebody needs something even faster, they'd do better forking NumDot to add their own functions. For them, #142 will be a better solution.