perf: improve performance of symbolic differentiation

[joshsunshine]

Description

Related issue/PR: #
#652

This PR dramatically improves the performance of the symbolic differentiation of diagrams with large numbers of constraints and objectives. Penrose was computing a gradient for each objective and constraint separately to provide debugging information. This PR removes that code and introduces a compressed graph representation for traversing the graph during symbolic differentiation. The graph representation can and should be used in optimization and evaluation as well. Perhaps it should be the output of the compiler.

Implementation strategy and design decisions

The compressed graph representation is based on this slide in Keenan's graphics course: http://15462.courses.cs.cmu.edu/fall2021/lecture/meshes/slide_021

Checklist

• I have commented my code, particularly in hard-to-understand areas
• My changes generate no new ESLint warnings

[codecov]

Codecov Report

Merging #840 (b30482c) into main (7d5538b) will increase coverage by 0.22%.
The diff coverage is 98.18%.

@@            Coverage Diff             @@
##             main     #840      +/-   ##
==========================================
+ Coverage   68.85%   69.08%   +0.22%     
==========================================
  Files          61       62       +1     
  Lines        7725     7776      +51     
  Branches     1454     1459       +5     
==========================================
+ Hits         5319     5372      +53     
+ Misses       2393     2391       -2     
  Partials       13       13              

Impacted Files	Coverage Δ
packages/core/src/engine/Autodiff.ts	74.33% <ø> (-0.10%)	⬇️
packages/core/src/utils/Graph.ts	98.14% <98.14%> (ø)
packages/core/src/index.ts	78.67% <100.00%> (+2.03%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7d5538b...b30482c. Read the comment docs.