[runtime] Reduce registered peer storage allocations - #12657

Reduces allocation overhead in the runtime’s registered-peer registry by inlining the first ReferenceTrackingHandle per identity-hash bucket and only allocating a list on actual collisions, while also adding BenchmarkDotNet benchmarks to measure registration/lookup/churn behavior.

Changes:

• Switch JavaMarshalRegisteredPeers storage from Dictionary<int, List<ReferenceTrackingHandle>> to Dictionary<int, RegisteredPeerBucket> and synchronize via a dedicated Lock.
• Factor peer replacement/keep logic into ReconcilePeer() and ensure struct-bucket mutations are written back into the dictionary.
• Add BenchmarkDotNet coverage for add/peek/remove-add churn scenarios across multiple storage strategies.

tests/Xamarin.Android.Tools.Benchmarks/RegisteredPeersBenchmarks.cs:311

• 💡 suggestion: This Peek method can return null even though it returns Peer. If a miss is unexpected in this benchmark, throwing makes failures explicit without changing the hot path.

	public Peer Peek (Peer expected)
	{
		if (!peers.TryGetValue (expected.Hash, out FirstRestBucket values))
			return null;

tests/Xamarin.Android.Tools.Benchmarks/RegisteredPeersBenchmarks.cs:414

• 💡 suggestion: Peek returns null on misses (including the single-handle fast path) even though the return type is Peer. Throwing on misses keeps the method contract non-null and surfaces setup errors clearly.

	public Peer Peek (Peer expected)
	{
		if (!peers.TryGetValue (expected.Hash, out object values))
			return null;

tests/Xamarin.Android.Tools.Benchmarks/RegisteredPeersBenchmarks.cs:476

• 💡 suggestion: This Peek implementation returns null on misses while returning Peer. If misses are not expected in the benchmark harness, throwing makes failures explicit and avoids a misleading non-null signature.

	public Peer Peek (Peer expected)
	{
		if (!peers.TryGetValue (expected.Hash, out InlineFirstRestBucket values))
			return null;