Optimize kernel performance

[ekuiter]

Some kernel profiling results on the client:

• FM-encode can be avoided by implementing the kernel.core.FM module completely in JavaScript
• functions that rely on the CDAG (e.g., apply-compatible* and CDAG/insert) can be optimized by using garbage collection (Garbage collection #30)
• preceding? is currently implemented with the traditional VC/_< comparator. Alternatively, a CDAG's CP can be used to obtain the same result in O(1). Question is, whether set access is actually faster than VC/< in practice.
• Improve performance of generate-operation!. Notably, we don't need to call MOVIC as we already know the outcome (just add the operation to the only CG).

(The 67ms from log are only added of concern in debug mode.)