Alignment dependency for AMO and uncached accesses

[sripathi-muralitharan]

I ran into the following issue while trying to develop a core<->L1.5 transducer -
For data that is returned from the L1.5 (i.e data0, data1, data2, data3), there seems to be a need to use the address at the transducer to pick out the data that is needed for atomic operations (LR/SC) and uncached accesses as opposed to the data replication that happens in case of a regular load or a store.

@fei-g @Jbalkind, more clarity on why it is the way it is and any nuances involved would be great!