Auto exclusions are broken

[jngrad]

There is a bug in the auto exclusions code that has always been present since the functionality was introduced in v2.2.0b by a48ed7c: the bond distance is interpreted as a particle id, because the loop counter is incremented by 1 instead of by 2. The 2D bookkeeping array is flattened in such a way that bond distances and ids are interleaved, thus requiring an increment by 2. We end up with spurious exclusions to particles with id 1 to n, with n the maximal bond distance requested by the user. If the particle list has a gap in the id range 1 to n, an error is raised.

MWE:

import espressomd
system = espressomd.System(box_l=[1.0, 1.0, 1.0])
bond = espressomd.interactions.Virtual()
system.bonded_inter.add(bond)
for pid in [0, 1, 2, 3]:  # use [0, 2, 3] to crash the simulation
    system.part.add(id=pid, pos=[0, 0, 0])
system.part.by_id(0).add_bond((bond, 3))
system.auto_exclusions(1)
for p in system.part.all():
    print(p.id, "->", p.exclusions)

Output:

0 -> [3 1]
1 -> [0]
2 -> []
3 -> [0]

Expected output:

0 -> [3]
1 -> []
2 -> []
3 -> [0]

This bug was never caught by the testsuite, because the test case only checks for linear topologies with contiguous particle ids starting from 0.