Databases from LGTM are built with odasa rather than codeql, and so don't have zipped source archives, and require an extra codeql database cleanup step to zip the source archive before jump-to-definition works on them.

It would be nice in principle if it just worked out of the box. With the current strategy of making one DefinitionProvider for the zipped archive scheme, it doesn't seem possible. We could perhaps, for each database add and delete, register and unregister a DefinitionProvider with an absolute file path pattern matching files in that database's source archive. Alternatively it might be the case that registering a broadly applicable DefinitionProvider isn't harmful as long as it simply returns zero results when another DefinitionProvider also exists.