extension for regression task

[itea1001]

added DefaultGenerationContinuous and DefaultUpdateContinuous
new reward: \frac{1}{1+MSE} + exploration
criteria for "wrong example": |pred - actual_value| > given threshold

[itea1001]

reward changed to:
$\frac{1}{n} \sum_{i=1}^n (reward_a - reward_b (y_i - \hat{y_i})^2) + \alpha \sqrt{\frac{\log t}{n}}$

(need to set $reward_a$ and $reward_b$ for DefaultGenerationContinuous and DefaultUpdateContinuous)