Apply babbaj's review of the Java pathfinder

[0Mattias]

babbaj reviewed the port on cabaletta/baritone#5117. His points were about the port's seams rather than its arithmetic, so they apply here unchanged, and this brings back the review commit from the fork's java-nether-pathfinder branch (0Mattias/baritone a8641949) — the one the pull request now carries.

• the port's own BlockPos is gone; it uses Minecraft's
• the int constants for the dimension and the cache-miss mode are enums, and hasChunkFromJava is hasChunkFromCaller
• the wrappers that existed to pack arguments for JNI — isVisible, isVisibleMulti, the batch raytrace, the long-packed path segment — are gone; callers use Raytracer.raytrace and List<BlockPos> directly
• Raytracer.raytrace returns the hit position or null instead of a boolean plus an out-array
• the search and the region reader time themselves with currentTimeMillis, and Chunk calls its 16-block slices sections
• BaritoneRegion checks CachedRegion.CACHED_REGION_MAGIC, now public
• elytraCustomAllocator is removed rather than kept as a deprecated no-op

Cheesecake's own differences survive it: NetherPathfinderContext keeps the UnusableRays guards and the read/write locking, so the per-ray calls that replace the batch wrappers sit inside them; the region reader still reads a region's blocks in the order Baritone writes them; package-info keeps its own wording. The pathfinder package is otherwise byte for byte the branch's, modulo the package name and the licence header, which is how it should stay while the pull request is open.

All 102 unit tests pass locally, including the oracle tests that hold the port to the native library's recorded answers, and CancelTest comes across with them.

🤖 Generated with Claude Code