Performance regression during upgrade from 2.88.0->2.89.0

[henri5]

DeepNesting.txt

Attached file from project which contains a lot of similar files, made generic to keep original structure and work with plain java classes. Such files are formatted on 2.88.0 in non-noticeable time, in 2.89.0 (and with its performance enhancements) and later it takes 10+ seconds, as a test case for FormatterIntegrationTest it takes 39 secs on my machine. With performance enhancement reverted, it takes 350ms.

In original project such extreme code blocks are marked as not to be formatted, but seems to take long time regardless.

[henri5]

Asked Claude to have a got at this and it essentially suggested (after few iterations) to restore branching coefficient check

State branchedState = this.state.withNewBranch();

if (branchedState.branchingCoefficient() < MAX_BRANCHING_COEFFICIENT) {
 // old code using branchedState as basis
}

around the main change in the breaking PR.