[ARBML/Tkseem] performance issues

[yousef337]

Problem:

Random trainer takes significant time due to its implementation, in addition, its implementations is unnecessary complicated.
The main applied method is _random_dict.

Solution:
Implement a reactive solution to increase the performance and code readability and useablity using pyrx library.

Note: It can be improved more by adding concurrency to down stream operations [map].
The new method is currently under testing