Stochastic-matching

* Extraction of metrics from simulation gathered in a unique submodule
* Most used metrics are now properties of the simulator object
* Pre-defined metrics can be selected by name on batch simulation
* Custom metrics can be used by passing their function

* Unified way to run batched of experiments
* Construct experiments with static and variable parameters
* Define how to extract the metrics you want
* Start evaluation and see how it progresses with tqdm
* mutiprocess.Pool can be optionally used to parallelize the results
* Results can be optionally automatically cached
* Cf notebooks or reference for details

* Changes in the simulator API:
* For k-filtering, the threshold parameter is now k
* weights are now called rewards everywhere but for priority (to keep the weight/counterweight story)
* Interleaving of rewards and forbidden edges has been improved (each can define the other if necessary)
* reward-based policies are triggered by setting a beta parameter
* Introduction of in-package parallelization tools
* New notebook tutorial added
* Bugfix: E-Filtering now has working CCDF
* Chores

* Simulator re-written almost entirely
* Easier to read/maintain thanks to jit and data classes.
* Roughly 40% faster than previous version.
* Virtual queue updated with better edge-FCFM policy.
* EGPD ported to both virtual queue and longest policies.
* epsilon-filtering (a.k.a. epsilon-coloring) added.
* Switch to Poetry
* Easier package maintainance
* Pydata documentation style
* Supported Python version: >=3.10

* Add a function to draw the (discrete) CCDFs piecewise
* New range of officially supported Python versions: 3.6 -> 3.11

* New optimize_rates for Model. Outputs a flow that optimizes the rates according to some reward weights.
* Refactoring: policies formerly called semi-greedy are now called (semi)-filtering.
* New option weights for filtering policies. Auto-computes the forbidden edges to optimize the reward according to weights.
* Default model tolerance raised to 1e-7 for better detection of null edges.
* Tutorials modified to introduce the new features.
* The notebook used for paper https://hal.archives-ouvertes.fr/hal-03502084 is now included in the documentation.
* Bug hunt: very large simulation could overflow silently (solved by switching logs from uint32 to uint64).