filters: add fixed-wander Kalman servo

[jordens]

This adds a single-estimator servo for clocks where oscillator wander can be known or bounded. Oscillator wander (in the model) is independent of the network environment.
The new servo keeps previous online network-noise estimation, but uses a characterized or conservative wander bound instead of a second estimator.

The default is four times the independently measured I210 wander; characterized targets should override it while unknown or
strongly varying oscillators should definitely keep the doubly-adaptive filter.

This assumption gives deterministic startup and lower CPU/memory cost.
Against the full doubly-adaptive KalmanFilter, the STM32H743 image
uses 2.8 KiB less text and 0.5 KiB less BSS. On the host, the full servo
workload is 1.54x faster. These numbers are on top of the gains in #871.

Note that this stacks on top of #871. And same as there:
I know statime is on hold for the big ntpd-rs work. I'm putting this out anyway because it may still become useful wherever the filters ends up living.

[davidv1992]

Apologies for the late response. Given the merger with ntpd-rs that is going on, and the direction we have chosen to take the algorithm there, this may not be a good fit for us then. It may be interesting once that work is a bit further along to convert this to the new algorithm traits over there, and keep an algorithm like this as an option in a separate repository.