Tracking issue for synthesis speedups

[ljedrz]

As requested in #2621 (comment) (which also contains some ideas).

Open PRs:

• [Perf] Reduce allocations in basic circuit operations #2620
• [Perf] Simplify the R1CS > Assignment conversion #2621
• [Perf] Use SmallVec in LinearCombination #2622

Potential future improvements:

1. use an arena (e.g. bumpalo) to create the vectors inside LinearCombinations
   • while perfectly possible, it would require all LinearCombination objects to carry a reference, which would require a major refactoring
   • this would obviate [Perf] Use SmallVec in LinearCombination #2622
2. use an arena to allocate the Rc objects
   • similar principle as 1, but not viable due to the allocator_api feature still being unstable; it might require less refactoring
3. while [Perf] Use SmallVec in LinearCombination #2622 is a solid improvement, a singlevec is slightly faster
4. once [Perf] Simplify the R1CS > Assignment conversion #2621 and [Perf] Use SmallVec in LinearCombination #2622 are in, this suggestion has good potential for a further improvement and simplification already refactored away
5. tweaking the Boolean bit operations can result in a modest improvement; they were reverted from [Perf] Reduce allocations in basic circuit operations #2620 in order to reduce the diff

What hasn't worked:

• using a leaner Arc (in tandem with [Perf] Simplify the R1CS > Assignment conversion #2621) which doesn't handle weak pointers

Cc @vicsn