Add hyper-parameter tuning with optuna

[juanmc2005]

This PR addresses issues #53 and #62.

Changelog

• Add diart.optim.Optimizer to tune hyper-parameters given a basic configuration object
  • By default, Optimizer creates a basic optuna study, but it can also work with a study created by the user
  • Optimizer is also compatible with optuna's distributed optimization
• Add HyperParameter data class to represent a tuneable hyper-parameter
• Add TauActive, RhoUpdate and DeltaNew instances of HyperParameter so they don't have to be created from scratch
• Add diart.tune script to quickly tune the default pipeline to a dataset (also compatible with distributed optimization)
• diart.stream, diart.benchmark and diart.tune can be run without python -m
• Make it easier to create a model block with SpeakerEmbedding.from_pyannote and SpeakerSegmentation.from_pyannote
• Separate verbose parameter in Benchmark into show_progress and show_report
• Make batch_size a constructor parameter in Benchmark
• Add menu to README
• Add custom embedding model example in README
• Add hyper-parameter tuning section to README