Hooking native method in oat image might cause crash.

[daddas]

When I was doing intergration of lsplant into libart.so. I found a tested hooking of simple jni function (defined in framework-classes) may crash the VM, especially when the java stack dump is performed.

It turns out that the jni method may have built-in oat header cur_oat_quick_method_header_ set, when doing StackVisitor::GetCurrentQuickFrameInfo().

This results in incorrect quick frame size for StackVisitor::WalkStack(bool include_transitions) process.

It can be fixed in art source code, while I think in Lsplant injection logics it may be inevitable.