Improvements for the search page

[karpov-sv]

Redesign of the search page to use "smart search" like in Google or ADS. Like the latter, offer the palette of search terms above the input field.

Possible queries are like:

• 10 20 300 - cone search at ra=10 dec=20 with 300 arcsec radius
• Vega r=10 - cone search within 10 arcsec around Vega
• 10 20 300 after="2023-01-01" window=1 - cone search within 300 arcsec from Vega in 24 hours after given moment
• ZTF22abzanvj - exact search for objectId
• ZTF23abshpmf r=100 - cone search within 100 arcsec around the alert
• ZTF22abzan - partial match search for objectIds starting with given pattern
• TRCK_20231212 - partial match search for tracklets (effectively, all tracklets from a given date)
• class=Unknown - last 100 alerts of a given class
• class=Unknown last=10 - last 10 alerts of a given class
• class=Unknown before="2023-01-01 01:12:23" - last 100 alerts before given moment
• class=Anomaly last=10 - last 10 anomalies
• after="2022-12-29 13:36:52" before="2023-12-29 13:46:52" - date range search with exact specification of start and stop times

Everything is somehow working, except for SSO - this one is waiting for backend API improvements as discussed with Julien.
GET api (does anyone actually use it?..) is changed to reflect underlying query parser, so the urls that work now look like:

• /?action=conesearch&ra=10&dec=20&r=300
• /?action=class&class=Unknown&last=10

Requires installation of two additional packages from pip - regex (used for partial matches) and dash-autocomplete-input (used for dropdown menu for class search).
Also requires updating dash-mantine-components to latest version from pip.

Documentation (mostly the help popup inside the search bar itself) is not yet updated.

[karpov-sv]

Card view added in latest commits actually quickly hits the limit of deploying with finite number of workers in Gunicorn (as suggested by README) due to its lazy loading of cutouts and lightcurves - as many component update requests actually perform one more request to /api/ under the hood, and block until it returned. So when there are many simultaneous ones the worker pool is exhausted and everything just hangs.

Solution might be to deploy a number of dedicated workers for /api/ that do not serve anything else. I did not find any built-in option in Gunicorn for that, so for now ended with deploying two instances on different ports, and pointing APIURL to the second one that is not visible from outside.

[karpov-sv]

With the last commit, the problem of exhaustion of web workers pool should be solved. The commit isolates all API access outside of /api/ into dedicated function that is able to either call requests.post(), or, if LOCALAPI=True, to use the endpoints directly, without doing unnecessary HTTP requests.

[karpov-sv]

At this point, I consider this PR feature-complete, and ready for review.

[JulienPeloton]

Thanks @karpov-sv ! I started to test all the new features -- it looks super nice! I noticed one big thing though: the switch card/table does not work. If I trigger the button, I get an error from the fact that the endpoint /api/v1/columns does not have the 'POST' method. But even if I manually fix this, the switch does not operate (no error though -- just doing nothing).

[karpov-sv]

Thanks @JulienPeloton for reporting, early morning coding has its downsides :-) I fixed it, should now be working.

[JulienPeloton]

Thanks @karpov-sv -- I tried all options, and everything works on my side :-)
Let's merge this!