Start the kernel ids over on reset.

So that things aren't affected by wrapping over.  Although, maybe we
should handle that better if it causes incorrect behavior...

Core/HLE/sceKernel.cpp

@@ -424,7 +424,7 @@ u32 sceKernelIcacheClearAll()
 KernelObjectPool::KernelObjectPool()
 {
 	memset(occupied, 0, sizeof(bool)*maxCount);
-	nextID = 16;
+	nextID = initialNextID;
 }
 
 SceUID KernelObjectPool::Create(KernelObject *obj, int rangeBottom, int rangeTop)
@@ -470,6 +470,7 @@ void KernelObjectPool::Clear()
 		occupied[i]=false;
 	}
 	memset(pool, 0, sizeof(KernelObject*)*maxCount);
+	nextID = initialNextID;
 }
 
 KernelObject *&KernelObjectPool::operator [](SceUID handle)

Core/HLE/sceKernel.h

@@ -427,7 +427,7 @@ public:
 	~KernelObjectPool() {}
 
 	// Allocates a UID within the range and inserts the object into the map.
-	SceUID Create(KernelObject *obj, int rangeBottom = 16, int rangeTop = 0x7fffffff);
+	SceUID Create(KernelObject *obj, int rangeBottom = initialNextID, int rangeTop = 0x7fffffff);
 
 	void DoState(PointerWrap &p);
 	static KernelObject *CreateByIDType(int type);
@@ -520,8 +520,9 @@ public:
 
 private:
 	enum {
-		maxCount=4096,
-		handleOffset=0x100
+		maxCount = 4096,
+		handleOffset = 0x100,
+		initialNextID = 0x10
 	};
 	KernelObject *pool[maxCount];
 	bool occupied[maxCount];