Fix issue with growing number of allocated direct buffers

[kochkozharov]

Example showing the problematic scenario is here in master branch: https://github.com/[kochkozharov/hadroNIO-memory-leak](https://github.com/kochkozharov/hadroNIO-memory-leak). It's pretty much the same as in my previous PR.
docker compose -f docker-compose.nofix.yml up --build
You can see unlimited growing number of off-heap buffers, which can cause OOM error or hang.
There are some charts from visualvm (shows memory usage and amount of direct buffers in taskmanager):

Same situation appears in jobmanager.
In this fix I suggest to implicitly free objects like new UnsafeBuffer(ByteBuffer.allocateDirect(Long.BYTES)) and replace direct buffer in callback with on-heap one.
Running the cluster with the changes:
docker compose up --build
Visualvm results in taskmanager:

The number of direct buffers has been significantly reduced.