CheriBoundAllocas pass should insert bounds between lifetime start/end intrinsics

[resistor]

Today the CheriBoundAllocas pass inserts bounds setting instructions between an AllocaInst and its lifetime markers. This is incompatible with new upstream behavior, insists lifetime intrinsics can only take AllocaInst's as arguments. This should be fixable by updating CheriBoundAllocas to insert the bounds-setting intrinsics between the lifetime start/end intrinsics instead.

[davidchisnall]

[ Per offline discussion ]

I think the right approach here might be for the IR pass to just do some annotation to mark allocas that are definitely accessed only in bounds and then to move the rest of the logic into the FrameIndex lowering.

I didn't do this originally because the MIPS frame index logic scared me.