More tuning of the redis session handler to reduce blocking wait time

A follow on from MDL-68577

We've hit a couple clients who have very large courses with hundreds of requests. These inevitably pile up and as soon as the page takes more than 5 seconds then each requests hits the hard coded limit and their session polling gets throttled back to once a second which means the images load roughly once a second each and block each other.

I'm not sure what to do here, the bottle next is really elsewhere and anything done here is a bandaid. However it could help things if when the 5 seconds is hit it slowly thottles back instead of a hard jump to every second. The original intention of the throttling was to avoid events which cause slowness from escalating and I want to retain that functionality.

So proposing something like we hit 5 seconds and then every poll after that we slowly ramp up the poll delay by 10% until we hit some cap.