Cache attachment context extracted by Tika

The Solr indexer currently extracts the textual content of attachments using Tika whenever an attachment is indexed. The same attachment is usually indexed at least twice: once on the document entry, once for a separate attachment entry, and then if the document has translations, also once for every translation. Whenever a document is changed, the attachment will be re-parsed even if it didn't change. As parsing attachments with Tika can be slow and resource-consuming, we should introduce a cache for the extracted text. This cache should ideally be stored in the same store as the attachment itself. There should be a way to manually or automatically clear the cache, e.g., on Tika upgrades.