Add L2 breakdown for FRONTEND_BOUND

[bcstrongx]

To match the Top-down method defined by Intel, FRONTEND_BOUND should have FRONTEND_BANDWIDTH and FRONTEND_LATENCY L2 events. FRONTEND_LATENCY accounts for the slots lost due to the latency of retrieving/decoding/issuing fetch data, in particular due to a TLB/cache miss. FRONTEND_BANDWIDTH refers to slots wasted due to decode inefficiencies, which I would only expect in cases where not all decoders are symmetric (e.g., some instructions require decoder 0, resulting in less than full decoder utilization in some cycles).