Changes for speed improvements in clustering jobs

[gartrog]

This PR should be merged at the same time as the one in k4FWCore.

Lots of small things that when put together allow to gain a lot.

• CaloTopoClusters: use unordered_map instead of map
• CaloTopoClusters: fix compilation warning
• CaloTopoClusterInputTool: optimize use of map: update values instead
  of clearing and filling again
• CaloTopoClusterInputTool: remove unused member variable
• CaloTowerTool: update to use the new interface (do not fill map of
  links if not needed)
• CaloTowerTool: micro-optimisation, reduce calls to eta and phi.
• CreateCaloClustersSlidingWindow: use new interface of CaloTowerTool to
  not fill map of links if attachCells is not used
• NoiseCaloCellsFromFileTool: use Decoder::get(ID, int) instead of string
• NoiseCaloCellsFromFileTool: simplify finding bin in histogram
• NoiseCaloCellsFromFileTool: micro-optimisation: avoid sqrt(x*x) when
  no pileup noise.
• SplitClusters: fix dangerous compilation warning
• CaloTopoClusterFCCee: use Decoder::get(ID, int) instead of string
• CaloTopoClusterFCCee: to avoid too many lookups in the huge noise map,
  precompute minimum noise per layer and start by comparing noise to
  this minimum noise before looking into the big map.
• CaloTopoClusterFCCee: micro-optimisation: avoid instantiating a
  vector<vector> of size 100000 at every call...
• CaloTopoClusterFCCee: fix compilation warning
• CreateCaloCellPositionsFCCee: create and use a cache of cell positions
  to avoid geomtry calls and conversions at every event
• CreateCaloCellPositionsFCCee: use Decoder::get(ID, int) instead of string

Note: many alogirithmic optimisations are certainly still possible, I
focused on easy gains that did not require deep understanding of the
code.
I suspect many calls related to geometry are far from optimal.

[gartrog]

Tagging @vvolkl and @BrieucF

[BrieucF]

Just tested it and everything looks ok to me!