Use the newly added completionItem/resolve request to compute
the 'documentation' and 'detail' fieds of completion items lazily
when the client supports it (i.e: when a given item gets selected
in the completion window).
This makes the ALS way faster to return completion items for packages
that contain a huge number of declarations (e.g: Libadalang.Common).
VS Code already supports this lazy computation method but not other
clients.
Support for tasks missing due to bug in libadalang [T716-049].
CalledBy request documentation typos fixed.
Add documentation for ALS Calls request.
Add a test for ALS Calls request.
A new custom request has been introduced to query the dependencies of a
given unit.
Documentation and a two tests have been added for this custom request.
Use a dedicated Unknown_Method exception. Replace if statements
in message decoders with Generic_Dispatching_Constructor and
a hash table.
Add test case for unknown request
Generate the boilerplate to support this, and implement the body.
Revisit the way we handle contexts: store the source files in an ordered
set to improve speed of lookup and memory management.
Make the indexing take place "in the background".
To do this,
* make the index subprogram stop itself when there is work on the queue
* add an interface that allows the server to do operations after each
request/operation, and use this to relaunch an indexing request
