Currently some checks are performed outside the update job, for example
the model is fetched and the javanature. These values can change between
submit the data to the job and it actually pick it up.

This now moves these check into the job itself and uses a concurrent
queue to prevent the need for synchronized. Also results are collected
inside a Map so if new update request are received in the meanwhile we
discard the previous computed values.

Beside that, entries are now eagerly computed as they are required
anyways when we set the classpath and we can catch problems that occur
here instead of just swallow them.

Also the job now react to cancel request much better than before.