Filtering outliers would indeed be very usefull. We'd need to check its impact on performance.
Regarding auto-restart : as ALS is not connected to the mount (and won't be in a near future) we would need to rely on repeated alignment failures to assume user slewed to a new target. Would be worth a test.
Please fee free to create feature requests as issues on ALS's repo at github.com/gehelem/als .Who knows, we might find some time or horsepower to put them into ALS