Reduce code redundancy

[gkjohnson]

For the sake of performance and avoiding additional if statements within tight loops there's a bit of code redundancy popping up throughout the code base, particularly with the new BVH super class #817. Some of the cases:

• Normalized vs non normalized attributes
• Indirect vs non indirect
• index vs no index
• interleaved vs non?
• Duplication of functions like "refit" for MeshBVH vs an abstracted, slower BVH version
• Duplication of computeTriangleBounds logic for MeshBVH performance

Currently we're relying on build-time generated functions to keep indirect and non-indirect code paths fast, which increases build sizes and will only get worse with more variants. Some options:

• Benchmark the performance differences between the custom MeshBVH computeTriangleBounds and the new abstract methods provided by BVH and determine whether the performance implications are acceptable.
• Use run-time code generation by basically implementing custom #define logic the way the build-time generation works.

[gkjohnson]

One issue with the run time code-gen is that if we use function.toString then it will likely have been mangled or minified by the build process, removing any ability to find any #define-like comments. And embedding all this to a string as a template will make builds larger than necessary.

[gkjohnson]

One solution would be to continue to write the code in "template" files and process it into a minified template object / builder on package build. Conditional blocks can be separated for integration later, imported functions can be passed as as context on build. Functions can be passed to the builder and inclined.