Faster bitwise-equal weld() implementation

[donmccurdy]

Provides an alternative weld() implementation based on Meshoptimizer's approach. When tolerance=0, we use a hashtable to find bitwise-equal vertices quickly. This is vastly faster than previous approaches, but comes without tolerance options. When tolerance is non-zero, weld() falls back on the older implementation. In v4.0, the new implementation will be the default.

Results

model	before	after
lovecraftian	1,232ms	149ms
plant_3	2,607ms	431ms
concessionaire	138ms	32ms
weld_problem	115ms	21ms

Context

• Fixes Improve performance of weld() function #738
• Related Improve weld() performance with spatial grid #803
• Related Performance improvements for large, complex models #1133

Tasks

• debug performance regression above. hash collisions maybe?
• add benchmark

[donmccurdy]

Current dependencies on/for this PR:

• main
  • PR Faster bitwise-equal weld() implementation #1237 👈
    • PR Shallow clone accessors when possible #1238

This stack of pull requests is managed by Graphite.