Refactor solution for question "Implement gradient descent variants with MSE loss"

[hoangble]

When working on it, I thought that the solution can be re-written a bit cleaner so I did.

1. Add type hints to parameters, variables, and function.
2. Refactor so that the only loop is written only once, and the batch_size is determined from the method parameter. This change highlights the differences among the variants.

It passed the test cases.

[hoangble]

@moe18 please take a look. Greatly appreciate!