Describe the feature or idea you want to propose

I propose the addition of a new forecaster class, TCNForecaster, under the module aeon.forecasting.deep_learning. This forecaster will use a Temporal Convolutional Network (TCN) for time series forecasting, built upon aeon's BaseDeepForecaster and a modular TCNNetwork class.
Refer #2931

Describe your proposed solution

The proposed implementation introduces a TCNForecaster class which:

• Inherits from BaseDeepForecaster
• Uses a configurable TCNNetwork to construct the underlying architecture
• Supports multi-variate input, customizable kernel size, dropout, and block depths

The forecaster is structured to follow aeon's deep learning forecaster pattern, enabling compatibility with the overall API and model selection tools.

Describe alternatives you've considered, if relevant

No response

Additional context

No response