CS_Shower::SetColours fails due to strange clustering

[Frank Siegert]

In the production of stable WZ+1j with MC@NLO I see a strange crash in current trunk from the shower which fails to set colours.

This is understandable in so far as the cluster history looks messed up:

    Cluster_Definitions_Base::ReCluster(0x17e5bf0) {
      (0x3eaadf0): 2 -> 4 {
        \mu_r = 90, \mu_f = 90, \mu_q = 90, \mu = 21.56
        k_T = 21.56, z = 0, phi = 0, kin = 0, K = 1.873
        oew = 2, oqcd = 3, nlo = 1, new = (5), ncl = 0, flag = 0
                 (0)           G (-105.1,0,0,-105.1) 1.349e-06 (0,601) [0|4]
                 (1)           G (-188.1,0,0,188.1) 3.815e-06 (0,0) [0|4]
                 (2)           Z (106.9,-10.05,43.48,-33.43) 91.19 (0,0) [0|4]
                 (3)          W+ (100.4,35.34,-48,8.153) 80.39 (0,0) [0|4]
                 (5)           s (13.57,-3.654,6.462,11.36) -1.686e-07 (601,0) [0|4]
                 (4)          cb (72.35,-21.63,-1.941,-68.99) 1.42 (0,0) [0|4]
      }
      (0x3ebc410): 2 -> 3 {
        \mu_r = 90, \mu_f = 90, \mu_q = 90, \mu = 16.37
        k_T = 36.33, z = 0, phi = 0, kin = 1, K = 1.873
        oew = 2, oqcd = 2, nlo = 5, new = (4), ncl = 0, flag = 0
               (0,5)           s (-92.07,-0,-0,-92.07) 1.349e-06 (0,0) [1|4] <-> (1)
                 (1)           G (-188.1,0,0,188.1) 3.815e-06 (0,0) [1|4]
                 (2)           Z (108.4,-11.47,45.99,-34.56) 91.19 (0,0) [1|4]
                 (3)          W+ (98.62,33.82,-45.33,8.12) 80.39 (0,0) [1|4]
                 (4)          cb (73.07,-22.36,-0.6613,-69.55) 1.42 (0,0) [1|4]
      }
      (0x3d60bb0): 2 -> 2 {
        \mu_r = 90, \mu_f = 90, \mu_q = 204.1, \mu = 204.1
        k_T = 204.1, z = 0, phi = 0, kin = 1, K = 1.873
        oew = 2, oqcd = 1, nlo = 1, new = (none), ncl = 0, flag = 0
               (0,5)           s (-92.07,-0,-0,-92.07) 1.349e-06 (0,0) [1|4]
               (1,4)           c (-113.2,-0,-0,113.1) 1.42 (0,0) [1|4] <-> (0,5)
                 (2)           Z (109.3,-23.04,45.65,-31.97) 91.19 (0,0) [1|4]
                 (3)          W+ (95.89,23.04,-45.65,10.9) 80.39 (0,0) [1|4]
      }
    }

Note that the "s cb" from the final state are clustered into initial state "s c", which violates the charges.

I am attaching the run card and debug output from one event. The crash is easily reproducible (typically after ~50 events) and the libraries+integration also don't take very long (a few minutes).