Add utilities for character level text generation

[kholdrex]

This pull request introduces a new text module and refactors the character-level LSTM example to use improved text utilities for vocabulary management, embedding, and sampling. It also adds a dedicated example demonstrating the use of these text utilities. The changes modernize the text generation pipeline, making it both simpler and more flexible.

Text utilities integration and example:

• Added a new text module to the library, providing TextVocabulary, CharacterEmbedding, and advanced sampling functions, and re-exported these for easy access. [1] [2]
• Created a new example file text_utils_example.rs that demonstrates the usage of vocabulary encoding/decoding, character embeddings, LSTM/linear pipeline, and various sampling strategies.

Refactor of character-level LSTM example:

• Refactored CharacterLSTM in text_generation_advanced.rs to use the new TextVocabulary and CharacterEmbedding, replacing manual character-index mapping and custom embedding logic.
• Updated training and generation logic to use cross-entropy loss and output logits over the vocabulary, with sampling handled by the new utilities.
• Simplified and modernized the code structure, removing redundant projection and sampling methods, and improving clarity in the training and generation workflow.