ekf optical flow only compensate to a lower value

[AlexKlimaj]

This PR is an attempt to remove fake velocity oscillations for the optical flow data. Where there is 0 or little pixel flow, the gyro noise oscillations dominate the optical flow velocity output.

Continuation of #22095

@dirksavage88

Current behavior in main.

Current behavior in main.

[dirksavage88]

Will test this very soon @AlexKlimaj

[bresch]

Let's assume you have a flow rate of 1rad/s and zero-mean gyro data with noise. If you simply subtract them, you'll get 1rad/s flow with noise. However, if you use your conditional subtraction, you'll get a signal with a mean < 1rad/s and some noise.

[dagar]

If we want to play with these ideas let's get a replayable log with GPS (or another reliable velocity/position source) at a few different altitudes.

[dirksavage88]

[AlexKlimaj]

Not sure if my conditional subtraction is working.

https://review.px4.io/plot_app?log=960fafa5-0818-4d50-96be-42134be18775

[dirksavage88]

I noticed the oscillations are far worse above 3m