Allow null-based ARR_ADDRs (#67486)

We can sometimes see trees like "ARR_ADDR(long 0)" that are the
result of morph's folding logic.

In general it does not seem great to allow folding of BYREFs
into LONGs like that, but it is also not incorrect per-se.

src/coreclr/jit/gentree.h

@@ -5788,7 +5788,7 @@ public:
         , m_elemType(elemType)
         , m_firstElemOffset(firstElemOffset)
     {
-        assert(addr->TypeIs(TYP_BYREF));
+        assert(addr->TypeIs(TYP_BYREF) || addr->IsIntegralConst(0));
         assert(((elemType == TYP_STRUCT) && (elemClassHandle != NO_CLASS_HANDLE)) ||
                (elemClassHandle == NO_CLASS_HANDLE));