Decouple the number of TDF slots from the number of workers in the IMT pool

Since the beginning, the number of processing slots for the TDF is identical to the number of workers in the IMT pool.
This works very well but may have limitations, most notably in presence of many workers and IO operations: the reading/writing from/to disk becomes the real bottleneck.
In order to aim to support HEP analysis @ 100 threads, it is necessary to decouple the concept of slot from the one of worker thread in the IMT pool.
The impact on the interfaces must be carefully considered.