Optimize deterministic scheduler performance

[ReubenBond]

Summary

• add a permanent BenchmarkDotNet scheduler-overhead benchmark with a trace mode
• eliminate per-step runnable-set, scheduling-context, transition-scope, and execution-context allocations
• combine allocation-free round-robin selection with active-operation tracking from latest main
• preserve custom/instrumented strategy snapshots and deterministic ordering

Performance

• baseline: 22.62 us and 372 B per scheduling point
• final: 20.13 us and 91 B per scheduling point
• improvement: 11.0% faster and 75.5% fewer managed allocations
• pvanalyze 2,500-run trace duration: 7.37s to 5.89s
• pvanalyze GC trace allocations: 157.04 MB to 64.35 MB

Validation

• 1,990 tests passed; 16 opt-in package smoke tests skipped
• BenchmarkDotNet default job with MemoryDiagnoser
• pvanalyze CPU and gc-verbose traces