CalorimeterIslandCluster: allow specifying adjacency matrix in terms of cellID fields

[veprbl]

The issue with the current implementation is that is_neighbour() relies on geometrical dimensions of a calorimeter being used. This makes it cumbersome and error-prone to enable clustering for different detectors. The goal is to instead use readout cellID fields.

This PR suggests a following interface for doing that: A new parameter adjacencyMatrix is added to describe the adjacency of the sensitive elements. It can be used replace the existing settings like localDistXY. The value specified must be a string expression (to be evaluated by DD4hep evaluator) that returns a "boolean" value which is non-zero for adjacent cells. An example predicate for a calorimeter with a readout ID spec like system:8,row:8,column:8 could look like

cl = IslandCluster("cl",
  readoutClass="EcalBarrelSciGlassHits",
  adjacencyMatrix="(abs(row_1 - row_2) + abs(column_1 - column_2)) == 1",
  # ...
)

A calorimeter that has sectors might require a more sophisticated predicate like:

(
  abs(tower_1 - tower_2)
  + abs((sector_1 - sector2) * 5 + row_1 - row2)
) == 1

There are few minor issues with the current implementation:

1. The 5 is a magical constant for number of rows per sector. It is possible to scan the geometry and define some sort of row_max constant to be used in place.
2. Expressions are evaluated for every tower. This may be slow, but those values can be precomputed if needed.
3. The DD4hep evaluator is designed to work with doubles, there might be some floating points caveats. Should not affect us unless we get into large numbers?

[veprbl]

Closing in favor of https://eicweb.phy.anl.gov/EIC/juggler/-/merge_requests/504