In the current implementation, every search opens/loads Whoosh index and initiates a new `Searcher` instance. This commit introduces a singleton to eliminate such behaviour, except when syncing  attachments that requires the new index. The downside of this change is that after syncing, the Flask app needs to be restarted for the searcher to use the new index.

In the current implementation, every search opens/loads Whoosh index to retrieve last sync time. This commit caches the last sync time. The downside of this change is that after syncing, the Flask app needs to be restarted for the searcher to use the new index.

This reverts commit acfcb32.

When server has small memory and documents are too many, syncing may cause OOM. This change may reduce such possibility.

Caching papers for 24hrs and reloading searcher index every 12hrs in case the index has been updated.

Compute the facet html in Python instead of in Jinja template. When facets are nested and in hundreds/thousands, this change can reduce rendering speed by hundreds of milliseconds.