L2rpn-baselines

-------------------------
- [BREAKING] remove support for gym, use gymnasium instead (if you still want
gym it should be fine to install `l2rpn-baselines` with `pip install l2rpn_baselines --no-deps`
and install gym elsewhere. But we do not recommend to do so)
- [BREAKING] change the signature of the "`GymEnvWithHeuristics.fix_action`"
- [FIXED] a "bug" due to the gymnasium / gym in grid2op
- [FIXED] way to retrieve the learning rate from Adam (in keras)
- [FIXED] `PPO_RLLIB` with new version of ray, rllib, gymnasium etc.
- [ADDED] example for training a model based on the "l2rpn_idf_2023" environment

------------------------
- [ADDED] the "topo oracle agent" (contrib)
- [ADDED] the "curriculumagent" (contrib)

---------------------------
- [FIXED] issue with the `PPO_SB3` agent when using a runner, particularly when no "heuristic" are
used at inference time.

--------------------
- [BREAKING] name of the file inside the submodule are now lowercase (PEP 8 compliance)
Use `from l2rpn_baselines.[BASELINENAME] import [BASELINENAME]` by replacing
`[BASELINENAME]` with ... the baseline name (*eg* `from l2rpn_baselines.DoNothing import DoNothing`)
- [FIXED] clean the documentation
- [FIXED] some bugs (especially in the type of actions) for some agents
- [ADDED] a code example to use stable baselines 3 (see `l2rpn_baselines.PPO_SB3`)
- [ADDED] a code example to use RLLIB (see `l2rpn_baselines.PPO_RLLIB`)
- [ADDED] an optimizer (see `l2rpn_baselines.OptimCVXPY`)
- [ADDED] some issue templates
- [ADDED] some examples in the "examples" folder

---------------------
- [FIXED] issue with grid2op version >= 1.2.3 for some baselines
- [FIXED] `Issue 26 <https://github.com/rte-france/l2rpn-baselines/issues/26>`_ : package can be installed even
if the requirement for some baselines is not met.
- [UPDATED] `Kaist` baselines
- [ADDED] The expert agent

--------------------
- [BREAKING] remove the SAC baseline that was not correct. For backward compatibility, its code
can still be accessed with SACOld
- [FIXED] the counting of the action types frequency in tensorboard (for some baselines)
- [FIXED] a broken Replay buffer `utils.ReplayBuffer` (used in some baselines)
- [FIXED] a bug in using multiple environments for some baselines
- [FIXED] wrong q value update for some baselines
- [IMPROVED] descriptions and computation of the tensorboard information (for some baselines)
- [IMPROVED] performance optimization for training and usage of some baselines
- [ADDED] better serializing as json of the `utils.NNParam` class
- [ADDED] the LeapNetEncoded baselines that uses a leap neural network (leap net) to create an
embedding of the state of the powergrid.