feat: add semantic TIGER recommendation module

[succSeeded]

Description

This PR adds a new rectools.semantic module that introduces semantic and generative recommendation tooling to RecTools.

• rectools.semantic.data_handling provides dataset preparation utilities for TIGER workflows, including sequence building, padding, iterative k-core filtering, and leave-one-out splitting.
• rectools.semantic.tokenizer provides semantic ID tokenization based on item embeddings, including tokenizer training, embedding datasets, quantizer abstractions, and RK-Means and RQ-VAE implementations.
• rectools.semantic.modules provides reusable neural network building blocks used by the semantic stack, including transformer layers, relative position bias, normalization layers, and MLP components.
• rectools.semantic.metrics provides ranking and recommendation diversity metrics used for semantic model evaluation.
• rectools.semantic.tiger provides the TIGER recommender itself, including the core network, training wrapper, loss computation, and high-level model interface.
• rectools.semantic.config_options defines typed configuration options for optimizers, quantizers, and learning-rate schedules used across the module.

Motivation

The module was implemented to bring semantic, neural recommendation workflows into RecTools in a reusable way. Rather than adding a standalone model only, this change introduces the surrounding infrastructure needed to prepare data, train the model, evaluate its outputs, and experiment with the approach inside the existing library.

Along with the TIGER implementation itself, the module includes shared semantic components, tokenization utilities, metrics, and example materials so the new functionality is easier to understand, test, and extend.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Optimization

How Has This Been Tested?

New tests were added to cover the introduced semantic module and its core components. The added coverage includes data handling utilities, tokenizer functionality, semantic metrics, reusable neural modules, and TIGER-specific logic to help verify that the new module integrates correctly with the existing codebase.