Reinforced-lib

Improvements:
- Update documentation.
- Add reference to the SoftwareX paper.

Fix:
- Normal Thompson sampling allows the `lam` parameter to be zero.
- Bernoulli Thompson sampling is stationary by default.
- Update the default value of the `decay` parameter the in `ra-sim` example.

Major API changes:
- Migrate from `haiku` (deprecated) to `flax` as the base naural network library.
- Update agent names to match literature:
- QLearning (deep Q-learning) -> DQN,
- DQN (deep double Q-learning) -> DDQN.
- Move particle filter from `agents` to `utils`.
- New behavior of loggers - all declared loggers get values from all sources.

New functionalities:
- Add Weights & Biases logger.

Other important changes:
- Fix updates with empty replay buffer.
- Fix logging of arrays to TensorBoard.
- Minor improvements in documentation.
- Rewrite Gymnasium integration example in documentation.
- Improve the CCOD example to better reflect the original implementation.

Improvements:
- Update documentation.
- Enable the use of 64-bit JAX.

New functionalities:
- Add the normal-gamma Thompson sampling agent.
- Add the log-normal Thompson sampling agent.

Fix:
- Make it easier to import the BasicMab extension.

Important changes:
- Move to `pyproject.toml` configuration file.
- Add basic extension for MABs.
- Update dependencies.
- Fix bug modifying user values passed to library functions.
- Fix agents behavior with multiple optimal actions - now agents draw one of the optimal actions instead of selecting the first one.