optimize ChunkProviderSuperPatcher: don't create ClassNode in shouldTransformClass

[danyadev]

Depends on FalsePattern/FalsePatternLib/pull/50
Will need to bump both compile and runtime dependency version after it's released

Now falsepattern takes only 3% of GTNH transforming time during client load, or only 150ms, which is 10 times less compared to initial measure:

For a reference, after only the first FalsePatternLib PR MergeableTurboTransformer.transform took 12.4ms for this particular class instead of 1ms now

Initial behavior: create a ClassNode, walk over all methods, then for every method walk over all instructions and search for CLASS_Chunk owner + METHOD_getBiomeArray name + ()[B descriptor

New behavior: first search for CLASS_Chunk in constant pool entries, then (if there's at least one match) search for METHOD_getBiomeArray in constant pool entries

It's a bit broader because we don't search for a descriptor, and method owner & method name are not necessarily connected to each other anymore, but potentially a bit larger subset of matched classes is still infinitely better than creating a ClassNode for every single class