Reduce ReaLearn's memory consumption

[helgoboss]

At the moment, there are two issues in the ReaLearn application/UI layer (not processing layer) that make ReaLearn consume much more memory than should be necessary.

• Issue 1 affects the memory consumption per mapping. This is related to the sheer amount of Rx subscriptions necessary to implement ReaLearn's totally reactive user interface. It will be tackled by partially moving away from Rx to a more scalable way of implementing the reactiveness.
• Issue 2 affects global memory consumption. It's related to the fact that ReaLearn loads all presets into memory at startup time. This issue could be less relevant when issue 1 is solved but it's probably still a good idea to load presets directly from disk and thereby save memory.

There are a few additional places that could be improved regarding memory usage but they are almost non-relevant compared to these two primary issues.

[helgoboss]

Turned out issue 2 is not that relevant anymore now that issue 1 is fixed. A good bunch of large presets now takes only 5 MB, which is much better than the previous 70 MB.