Regression in schema compilation memory requirements

I have a fairly large schema that compiles successfully on 3.1.0 with 200MB of memory given to the JVM.

With the current main branch, I have to bump the memory up to 1300MB before it compiles, otherwise I get an OutOfMemoryException. This is about a 7x bump in memory.

Git bisect pinpointed the regression to commit 62ac1047a18979bea1f5d6c668eefeaaae39d64b

Skimming that commit, nothing immediately jumps out at me as a significant change to what we already did, so I'm not sure exactly what caused this regression.