Summary

Fix ZScheduler worker-state accounting so the scheduler doesn’t treat the searching count (low 16 bits) as the active worker count.

This prevents workers from entering “searching” mode too aggressively, reducing unnecessary state churn (and contributes to less park/unpark thrashing).

Change

• Correctly extract active count from the high 16 bits of state.

Proof

Tests:

sbt -Dsbt.supershell=false 'coreTestsJVM/testOnly zio.internal.ZSchedulerSpecJVM'

See attached output in proof_9878_tests.txt.

/claim #9878