Implement new padding manager calculation algorithm

[lodoyun]

Description

This PR implements a new algorithm to decide the amount of padding to send.
The new algorithm is based on a FSM with 4 states:

• Start - a ramp-up state where we will send more padding than the current estimate. This state will be triggered when a stream starts and will last for a few seconds.
• Stable - a mode similar to the old algorithm that adjust padding based on the current estimate, the target bitrate and the media that is being sent.
• Hold - After a big drop in estimate is detected, we will go into hold mode and stop sending padding for a few seconds.
• Recover - After being on hold, this state will check if the estimate drop was transitory by adjusting padding with the estimate detected before the drop. If another drop is detected in this state, the new lower estimation will be used in stable mode.

We currently have no way to "restart" which may be a good idea in some situations (big changes in the target bw, for instance)

• It needs and includes Unit Tests

Changes in Client or Server public APIs

[] It includes documentation for these changes in /doc.