Fix ResourceLoader deadlocks

[hpvb]

This fixes several issues:

• There was a bug in the cycle detection. It was possible for the cycle detection to fail because the current thread isn't in the list. Fix this by just seeing if any of the next threads were already in the cycle.
• By way of an optimization the thread_waiting_on bookkeeping was done under the same thread_load_mutex as the cycle detection. But there is also an early exit after yields. If we exit through the yield the thread_waiting_on entry is stale and can cause other threads to fail to do cycle detection.
• When multiple threads end up finishing a resource simultaneously anyway, probably through cycle detection, there's a small chance that the original thread finished more-or-less at the same time if the original thing blocking load completed. We solve this now by letting only one of the threads do the initial registration.

Testing was done with MRPs from:

• ResourceLoader.load_threaded_request() crashes the engine if called for multiple complicated scenes #111202
• Crash/missing resources on load_threaded_request with use_sub_threads #115069
• LoadThreadedRequest can crash when loading the same subresource as another thread #118085
• Non-deterministic crash in resource loading on game launch from editor #119371
• Resources loaded using ResourceLoader.load_threaded_get are never unloaded #119374
• ResourceLoader.load_threaded_request use_sub_threads = true causes Player to crash semi-randomly #84012
• Using load_threaded_request with sub_threads raises errors when requesting multiple scenes with scripts that access the same autoloaded node #98865

In addition two projects that were shared with me privately.

No regressions compared to current master were found, and the new issues were fixed.

[bruvzg]

Seems to fix the only crash I was still able to reproduce. And none of the previously fixed crashes are reproducible.

[Repiteo]

Thanks!