Fix bug in accumulation of blocked nodes

The old implementation was not correct on complicated trees. If a tree was a bit heterogeneous, then the parent-child reduction would take increase the result of one sub-tree to the starting count of the next sub-tree. Thus having higher blocking than what is supposed to be.
2 jobs for support/dune-copasi-latest in 7 minutes and 12 seconds (queued for 2 seconds)
Status Name Job ID Coverage
  Test
failed debian:10--clang #244048
cores:4 mem:16G

00:05:42

failed debian:10--gcc #244047
cores:4 mem:16G

00:07:12

 
Name Stage Failure
failed
debian:10--clang Test
--- Failed to build dune-pdelab ---
Terminating dunecontrol due to previous errors!
+ exit 1
Uploading artifacts for failed job
Uploading artifacts...
junit/*.xml: found 1 matching files and directories

Uploading artifacts as "junit" to coordinator... ok
id=244048 responseStatus=201 Created token=Ys1eJwHx
Cleaning up file based variables
ERROR: Job failed: exit code 1
failed
debian:10--gcc Test
--- Failed to build dune-pdelab ---
Terminating dunecontrol due to previous errors!
+ exit 1
Uploading artifacts for failed job
Uploading artifacts...
junit/*.xml: found 1 matching files and directories

Uploading artifacts as "junit" to coordinator... ok
id=244047 responseStatus=201 Created token=aSproaJd
Cleaning up file based variables
ERROR: Job failed: exit code 1