LArRodDecoder is also cached in the HLTCalo infrastructure to avoid having to...

LArRodDecoder is also cached in the HLTCalo infrastructure to avoid having to find it again at each call - and avoid making request to find context many times. This MR is fundamentally important to complete the migration to MT.It mas look confusing, but it is exactly the best option we have now. I would urge it to pass as soon as possible.