A previous fix for SPIGOT-6814 implemented a callback function for the
PDC implementation that could be set to actively define a chunk as
unsaved, allowing chunks that have not been mutated through block
changes to still require saving if the chunks pdc was mutated.
This implementation however would pass a callback that references the
chunk access internally, meaning the PDC now actively holds onto a
callback that holds a reference to the entire chunk.
Aditionally, this change also impacted the pdc for item metas and
entities for really no reason whatsoever.
This commit re-implements the fix by introducing a new child of the pdc
implementation that the chunk now uses as its pdc. This specific
implementation maintains a dirty flag that is set to `true` on any form
of mutation and set back to false by the chunk that owns the PDC
whenever the chunk itself is flag as no longer dirty.
* CraftBlockState#getWorldHandle() would previously run into a NPE when being invoked for a non-placed BlockState. It now returns null in this case.
* CraftChest#getInventory() would previously encounter this NPE when being called for a non-placed BlockState. It will now automatically forward the call to #getBlockInventory() when it is being called for either a non-placed BlockState, or during world generation.
* CraftStructureBlock#applyTo was able to run into a NPE when being called for non-placed BlockStates (this is for example called by #getSnapshotNBT()).
* Replaced all no-world-generation preconditions with a general 'ensureNoWorldGeneration' method.
* Removed a few superfluous #isPlaced() calls. If getWorldHandle() reutrns a World, this already implies that the BlockState is placed.
During block destruction, the type of the block may already have been set to AIR while the TileEntity has not yet been removed.
Also, TileEntity#getOwner() skips the whole BlockState construction now if the block is of type AIR.
This removes the previous workaround again of returning a dummy CraftBlockEntityState in this case.
