Investigate obviously incorrect guide corrections from AG

Cilpped email from Akira Arai

...

Under unfavorable sky conditions, the AG guider occasionally produces
large outlier values in RA and Dec errors. Currently, when autoguiding
is enabled, the AG actor appears to send all calculated error values
directly to the telescope, without rejecting these outliers.
This can sometimes result in incorrect telescope movements with large
pointing offsets.

Under normal sky conditions, RA and Dec guiding errors are typically
within 0.5 arcseconds (with an average of ~0.2 arcsec). However, when
clouds pass through the field, the number of stars available for
fitting temporarily decreases, and the RA/Dec errors can exceed 10 or
even 20 arcseconds. This may cause sudden jumps in telescope pointing
and lead to the loss of guiding stars.

Therefore, I would like to request the implementation of an option to
reject large outlier values—that is, to prevent them from being sent
to the telescope.

In the meantime, it would be helpful if support astronomers could set
an upper limit for accepted guiding errors using the "autoguide
reconfigure" command. This functionality is important for maintaining
stable instrument operation, and early implementation would be greatly
appreciated.

Looking ahead, it may also be beneficial to implement more robust
filtering methods, such as sigma clipping, based on error statistics
within the same visit.

...

Tasks:

• Remove obviously large outliers
• Allow for "autoguide reconfigure"
• Build a guide history for particular visit and refine guiding based of knowledge of current visit conditions.

yuki.moritani says large error in visit=126457 agc_exposure_id ~ 726470