Reduce allocation pressure

[MoePus]

This PR reduces allocation pressure in FlowSched hot paths.

Changes:

• Avoid allocating FreeableTaskList for tasks without lock tokens.
• Reuse a shared empty LockToken[] for SimpleTask.
• Lazily allocate per-status ticket sets in TicketSet.

In my worldgen stress test, JFR showed ExecutorManager.tryLock() / FreeableTaskList as a major allocation source before this change. After the patch, that allocation site disappeared from the top allocation sites.

Observed result in the my test setup:

• allocation rate dropped from 6 GB/s to 3.5 GB/s
• GC intervals became noticeably longer
• per-worker FlowSched/C2ME allocation pressure was significantly reduced in JFR

This pr was coworked with Codex
This pr is the re-targeted version of #2