Make version semvar compliant
Removes warning log from startup

Init emptiness map lazily if needed
Fixes #56

Fix large fill commands breaking light (Vanilla)
Vanilla didn't tell the client to recalculate the edges of the
light it checked, and nor did it even tell the client
to perform the light updates. So of course it broke on chunk edges.

Fix is to simply send the neighbouring light data as well.

Fixes #71

Relocate save/load hooks into their own class
This will make compatibility with starlight better in the future for
mods that mess around with chunk NBT.

There is no forge version for 1.17 yet, so there is no RC3 for forge. No, I will not backport RC3 to 1.16.

Technical details document
Can be read here: https://github.com/Spottedleaf/Starlight/blob/fabric/TECHNICAL_DETAILS.md
Please note that document is intended for programmers.

Correctly decrease light when block data section is de-initialised
Fixes broken lighting on empty chunk edges when a neighbouring section becomes empty
by removing a source block. It was a rare issue, so it was never reported to me - I found it by
accident.

Handle concurrent chunk generation and lighting better
Thanks to the top tier chunk system, chunk generation
and lighting can occur at the same time. To fix this,
I've adjusted the code to handle concurrent updates
after the chunk is lit, similar to how Vanilla would have handled
it.

Update README to include new benchmarks and other revisions
Please note that these new benchmarks are taken on a newer CPU than
the one I used for the old benchmarks. So do not compare the new ones
and the old ones, you cannot compare benchmarks running entirely different hardware.

Completely rewrite how lighting is loaded in for clients
Attempts to fix #25 and #40
The old hook was vulnerable to mods doing unexpected things, especially
serverside. The old method also relied on unreliable ordering of events
inside packet handling. New method just hooks directly into the
packet handling methods to avoid conflicting behavior with mods and
to create more reliable hooks.

Completely rewrites internal scheduling of light tasks
This is the fix for chunk loading breaking while under
high block update stress. I know there was an earlier commit,
but hindsight is telling me that commit really didn't do anything.
However the complete rewrite of the internal scheduling will
fix the problem.
Old scheduling was very vulnerable to block updates completely
overloading everything. The old scheduling was also incapable of
indicating to the chunk system whether block updates or other
light modifying tasks were pending for a chunk. The new
scheduling now groups tasks by chunk (light update, needs lighting,
edge checks) and simply drains tasks by chunks. In the event
that there are a lot of block updates, they no longer pile up
in a queue since they are just grouped onto their chunk, which
will eliminate duplicate updates. And since it's a FIFO by chunk,
the maximum latency for a light update to happen is simply the
number of chunks queued. The new queue also tells the chunk
system to not unload chunks until any lighting task is fully complete
for that chunk and its 1 radius neighbours. This will eliminate
problems where the light engine was never able to complete tasks
before server shutdown. This also officially kills light suppression.
Before you ask, no I'm not going to re-add lighting bugs to my light
engine. I want lighting to work, which means I need to fix it when
it clearly doesn't work.

Mark the vanilla light engine fields in LevelLightEngine as null
This will break any mod expecting them to be non-null. But since
we replace the light engine, it's better this way since it forces
an explicit break rather than a silent break.

Initialise block nibbles in 1 radius of non-empty chunks
While vanilla is perfectly capable of handling null nibbles
in these cases, some protocol hacks might not expect this behavior.

I've taken an approach to mod conflicts that there's really nothing
I can do about it since I rewrite the light engine. Reporting them is still
useful for other people to track what is and isn't working however. Sometimes
I can do things about them, but I wont lose sleep if I can't.

Most of the remaining problems I see are with mod conflicts. Can't really do anything about them,
so they're not going to block full release.
There do appear to be some problems with how the client is receiving light data, and hopefully
the fixes in 1.0.0-RC1 fix those issues.

• Significantly reduce memory overhead    

• Rewrite nibble initit    

  This fixes all of the known issues around nibble init     

• Fix save hook not firing    

  Now we will no longer be relighting all of the chunks that we saved 
  

Move to edge checks for loading and generating light 

While it might appear at first glance that the vanillachunk system makes a guarantee that 1 radius chunks areloaded for lighting, this isn't actually the case. In fact,no chunks could be loaded!

And yet, the light function isactually required to be completed still! The only waywe can do this is with edge checks. 

In the 1.17 snapshots, it looks like chunk loading doesn'tload 1 radius chunks - which means our logic to require 1 radiusjust has to go. In order to ensure that this new behaviordoesn't cause lighting problems on chunk edges, chunk loadingnow does edge checks. This fixes some rare light engine crashes. 

1.17 snapshot version

1.17 should work just fine, the amount of changes required to update were basically minimal. Please note when testing 1.17 that the maximum world difference from top to bottom is 512 blocks - there are technical problems in the vanilla snapshots preventing a larger difference from being used, and attempting to do so will result in totally broken clientside chunk loading.

•  First beta release of the engine. Please note that since this is a beta, there could be some severe light corrupting issues. Known issues:     

• The Starlight fabric mod, unlike the Tuinity implementation does not have general improvements around the chunk system and shape math that Tuinity/Paper has, which need to be ported. This means the fabric mod might not live up to the performance of the generation results, as those were obtained on the Tuinity Starlight implementation. Notably, lighting snowy areas is notably worse on the fabric mod compared to the Tuinity implementation.     

• There are lighting issues on the client that have yet to be resolved
  

1. Fix chunk relightingIn order to make relighting more reliable, move theemptiness map lookups to their own cache. This allows usto not clobber/use emptiness map values for chunks that werenot relight during the call.More importantly, fix a typo where we supplied the wrong chunkargument to the emptiness change function for chunk neighbours -this caused the nibbles for the target chunk to be clobbered.

2.  Fix broken edge checksWhile edge checks are not currently used, it doesn't meanthey should be broken. Edge checks should be able to happenon null nibbles for block lighting (sky lighting will not havenibbles marked null during edge checks). 

3. Make handleEmptySectionChanges non-relaxed serversideFor the server, it should be that the chunks in 1 radius areloaded during this call. 

4. Fix incorrect neighbour nibble init logicWe need to be copying the whole array, not justthe first 9 values.  

5. Fix potential missed updateVisible callsrewriteNibbleCacheForSkylight could set a valuein the cache to null and not update the nibble.Now the call will call updateVisible whenremoving nibbles from the cache. 

6. Fix skylight propagation on the -1 chunk sectionAllow sources to be set up on y=-16, and fix lightretrieval for y < -16