Lvmguider

🔧 Fixed

* Actually import the `corrections` command.
* Use WCS to determine `RAMEAS`, `DECMEAS`, `PAMEAS` in co-added images.
* Wrap `offset_pa` in the -180 to 180 range.

In addition to the changes listed under `0.4.0b2`, this version provides:

🚀 New

* [8](https://github.com/sdss/lvmguider/pull/8) Added `corrections` command.

🔧 Fixed

* Reported FWHM were actually Gaussian sigmas. This has not been fixed.
* Improved matching of extracted sources with Gaia DR3.

In addition to the changes listed under `0.4.0b1`, this version provides:

🚀 New

* [7](https://github.com/sdss/lvmguider/pull/7) `stop` command now cancels the guider task for faster performance.

✨ Improved

* If a single camera fails to match with Gaia during guiding, retry that camera using astrometry.net.
* Use 25% percentile to calculate the FWHM.

🔧 Fixed

* The co-add file watcher seemed to skip files. Apparently this happens because some events are detected as a file being moved (from a temporary file to the final `.fits.gz` file) instead of a file creation. This version handles both kinds of events.
* Various fixes for the co-added code when dealing with missing data.

🚀 New

* [5](https://github.com/sdss/lvmguider/pull/5) Major refactor of the guider code.
* Extracted sources are always matched to Gaia DR3 regardless of whether the solution was obtained using astrometry.net or kd-tree. `lmag` zero points are calculated for each source.
* Once acquisition is complete the WCS of the individual cameras is determined from the Gaia cross-match, and not by tranlating the reference WCS using the measured offset. This allows each new WCS to also reflect changes in PA.
* Average zero points and PAs are output.
* All the metadata is carried out in two dataclasess, `CameraSolution` and `GuiderSolution`.
* The output data model has changed. Former `proc-` files are now named `lvm.{telescope}.guider`, and sources are saved as `parquet` tables. The latter allows to preserve the column types in the presence of missing data.
* [6](https://github.com/sdss/lvmguider/pull/6) Major improvements to co-adding code. QA plots generated during the co-adding process and the results are loaded to the database.
* Implemented guiding in rotation/PA.
* Added script to reprocess all MJDs at LCO and generate co-added products.
* Added file watcher to generate co-adds during the night. Updated the container image to support running the file watcher as a Kubernetes service.

✨ Improved

* Improved the sigma clipping of FWHM outliers.

🏷️ Changed

* Renamed references to "master frame" to "full frame". Coordinates are now referred to as `x_ff` and `y_ff`.

🚀 New

* [3](https://github.com/sdss/lvmguider/pull/3) Plotting of focus sweep and fit data using a cubic spline.
* [4](https://github.com/sdss/lvmguider/pull/4) Added code to generate co-added frames and QA metrics. This code will probably change significantly before it's used routinely.

✨ Improved

* Command telescope offsets with more relaxed thresholds for axis errors. Timeout and continue guiding if the offset takes longer than a certain amount of time.
* Command PlaneWave offsets with more relaxed `axis_error` and a reasonable timeout to prevent offsets taking long to complete under windy conditions.

✨ Improved

* Use actor model for `lvm.TEL.foc` to reduce polling.
* Round up `focus_position`.
* Prevent `astropy` from downloading data from the internet.

🔧 Fixed

* Set guider to `IDLE`` after` `stop --now`.