perf: replace global lock in EventReceiverRegistry with lock-free structure

[thomhurst]

Summary

EventReceiverRegistry.RegisterReceivers takes a global Lock once per test. With Environment.ProcessorCount * 4 concurrent tests this is a top source of contention in the trace: Monitor.Enter_Slowpath at 2.16% self. Estimated ~1.5% CPU + removes hot lock.

Evidence

Trace top exclusive:

7.  Monitor.Enter_Slowpath(...)                   2.16%  (self)

Code: TUnit.Engine/Events/EventReceiverRegistry.cs:29-96, 145-168. RegisterReceivers is called from TUnit.Engine/Services/TestExecution/TestCoordinator.cs:67 on every test start.

The registry is conceptually a copy-on-write Dictionary<Type, object[]> — currently guarded by a single shared lock.

Proposed fix

• Switch to ConcurrentDictionary<Type, ImmutableArray<object>>.
• Update via AddOrUpdate with ImmutableArray.Add or a snapshot-and-swap via Interlocked.CompareExchange.
• Short-circuit RegisterReceivers entirely when every object in the incoming span is already in _initializedObjects (common case — most eligible objects are shared across tests).

Expected impact

• ~1.5% CPU reduction on parallel paths.
• Eliminates lock contention that scales with parallelism (gets worse on bigger machines / bigger suites).