Reworks model loading almost entirely to deal with the new model-loader-api-v1 in fabric and the sudden no-deprecation removal of v0 from dev artifacts.

The new loader:

• Grabs all its raw data ahead of time on a DataLoader worker thread - this is extremely fast, usually on the order of 300-400msec on a vanilla instance, but times are posted in the logs now.
• Builds and prunes model trees in a preparation worker thread. Typically around 400-500msec for the whole kit, this time is also posted in the log. This is when memory usage peaks, but big data structures are nulled out ASAP for the garbage collector in case things get hairy.
• Builds / provides / bakes the final model in yet another worker thread. The main speed benefit is here, that by the time the model is requested by Id, the data is already present and we don't need to touch ResourceLoader at all.

The speed boost in loading individual models comes at the expense of a moderate memory spike and some very particular setup steps. Please let me know if either of these gets out of hand.