Granary Dataset Processing (Component-Based)

[ssh-meister]

This pull request introduces a set of modular components for processing the Granary dataset.

🔧 General-Purpose Processors:

These processors are not specific to any single dataset and can be reused across different data pipelines:

1. LambdaExpression processor LambdaExpression processor implemetation #136
2. SubRegex processor: adds support for extracting a list of regex parameters from a YAML file SubRegex processor: substitution rules from an external YAML  #137
3. ExtractTar, RemoveFiles processors Add RemoveFiles and ExtractTar, reorganize audio converters #139
4. FasterWhisperInference, DetectWhisperHallucinationFeatures, vLLMInference and CleanQwenGeneration Refactor inference processes & add new engines (FasterWhisper, vLLM) #141
5. ListToEntries processor ListToEntries processor #140
6. DropSpecifiedFields processor DropSpecifiedFields processor implemetation  #144
7. CharacterHistogramLangValidator processor CharacterHistogramLangValidator processor implementation #154
8. FastTextLangIdClassifier processor FastTextLangIdClassifier processor implementation #149
9. CometoidWMTQualityEstimation processor CometoidWMTQualityEstimation processor implementation #151
10. ConvertToTarredAudioDataset processor ConvertToTarredAudioDataset processor implemetation #145

⛓️ Pipelines

1. Unified pipeline and README with instructions and documentation Granary large-scale speech processing pipeline  #155