seems it could deadlock in rare cases somehow

Plan:

Processor reads nbt block. If dry and waterloggable, store blockpos into a WorldSavedData.

WorkdSavedData has a map of chunkpos to set of blockpos so blockpos are grouped by chunkpos. Is saved and read from nbt to prevent data loss. Every tick, it checks if the chunk is fully created and if so, then it reads the blockpos for the chunk, gets the block from the chunk, if it is a waterlogged block, set waterlogged to false and set the block in the chunk.