The object type loading system was refactored, resulting in the following key improvements:

1. Efficient parallel execution: Instead of launching multiple threads with manual task distribution, Parallel.ForEach with Partitioner is now used, ensuring automatic load balancing and optimal utilization of CPU cores.
2. Lazy JSON resolution: The need for expensive DeepClone() for each object variant has been eliminated. Instead, a static Resolve method has been introduced that processes only those parts of the JSON that require modification, significantly reducing memory consumption and speeding up variant processing.
3. Improved code structure: The logic is divided into clear steps-loading base types, processing them in parallel, and final registration. This improves readability and makes the code easier to maintain.
4. Reliable thread management: The _gatheringCompleted flag has been introduced, eliminating fragile wait loops via Thread.Sleep, making execution more predictable and efficient.
5. Enhanced error handling: errors in processing certain types no longer interrupt the entire process, but are isolated and logged, ensuring that other objects continue loading.

Also this PR is important here.

This PR also comes as a bonus.

Before (1.21.6):

After: