[Feat] : Add gradient clipping utilities for neural network training

[Advaitgaur004]

This PR introduces essential gradient clipping functionality to help stabilise training and prevent gradient explosion

New Functions:

• cten_clip_grad_norm() - Clips gradients by global norm (L2)

// Clip gradients by their overall magnitude
void cten_clip_grad_norm(Tensor* params, int n_params, float max_norm);

• cten_clip_grad_value() - Clips gradients to symmetric range [-max, max]

// Clip gradients to stay within [-max_value, max_value] 
void cten_clip_grad_value(Tensor* params, int n_params, float max_value);

• cten_clip_grad_value_range() - Clips gradients to custom [min, max] range

// Clip gradients to custom range [min_value, max_value]
void cten_clip_grad_value_range(Tensor* params, int n_params, float min_value, float max_value);

• cten_clip_grad_positive() - Clips only positive gradients above threshold

// Only clip gradients that are too positive (> max_value)
void cten_clip_grad_positive(Tensor* params, int n_params, float max_value);

• cten_clip_grad_negative() - Clips only negative gradients below the threshold

// Only clip gradients that are too negative (< min_value)  
void cten_clip_grad_negative(Tensor* params, int n_params, float min_value);