Cache pipe shapes by connection mask

[Marsietiss]

What

Caches the combined base VoxelShape returned by PipeBlock#getShapes for each possible connection mask.

Implementation Details

Adds a lazily populated 64 entry AtomicReferenceArray<VoxelShape> to each PipeBlock. The six connection bits are used as the cache index and compare-and-set safely publishes shapes when client and server threads query the same block at the same time.

AI Usage

• No AI driven tools were used for this pull request.
• Yes AI driven tools were used for this pull request.

Outcome

Removes repeated construction of identical pipe shapes during collision queries.

• Aggregate time for the affected entities dropped from 12.9% to 3.2%.
• Average per-entity tick time dropped from approximately 344 μs to 64 μs.

Before: https://observable.tas.sh/p/kRjSS/aggregated
After: https://observable.tas.sh/p/SidOa/aggregated

How Was This Tested

• Ran ./gradlew spotlessCheck build against the current 1.20.1 branch.
• Loaded the same world with the patched GTCEu 7.5.3 jar and captured the profiles.

Potential Compatibility Issues

Should be none

[DilithiumThoride]

Was this cache tested with pipes of different sizes but the same connections? i.e. if a tiny pipe connected N/S caches a collision box, will a huge pipe connected N/S pull the same collision box? or will it generate a new collision box for itself?

sorry i thought PipeBlock had only 1 instance per pipe size

[screret]

LGTM