Skip validation for ddocs in the scanner

Previously, if users had invalid ddocs in the cluster, for instance something
like `_design/`, the scanner would crash when tranforming them into `#doc{}`
records. That happened because the `couch_doc` function would automatically
revalidate them: check max doc length ID, attachments format, revision format,
doc ID format, etc. Which altogether silly, since they are already saved in our
cluster.

So, to avoid wasting resources revalidating existing documents, and crashing
the scanner process, use a much simpler function to transform the ddocs: make a
new `#doc{}` record with the ID and body by throwing away all the "special" doc
fields we don't care about (`_rev`, `_conflicts`, ...).

Author: nickva

src/couch_scanner/src/couch_scanner_plugin.erl

@@ -329,7 +329,7 @@ scan_ddocs_fold({meta, _}, #st{} = Acc) ->
     {ok, Acc};
 scan_ddocs_fold({row, RowProps}, #st{} = Acc) ->
     DDoc = couch_util:get_value(doc, RowProps),
-    scan_ddoc(couch_doc:from_json_obj(DDoc), Acc);
+    scan_ddoc(ejson_to_doc(DDoc), Acc);
 scan_ddocs_fold(complete, #st{} = Acc) ->
     {ok, Acc};
 scan_ddocs_fold({error, Error}, _Acc) ->
@@ -633,6 +633,13 @@ fold_ddocs(Fun, #st{dbname = DbName} = Acc) ->
             Acc
     end.
 
+% Simple ejson to #doc{} function to avoid all the extra validation in from_json_obj/1.
+% We just got these docs from the cluster, they are already saved on disk.
+ejson_to_doc({[_ | _] = Props}) ->
+    {value, {_, DocId}, Props1} = lists:keytake(<<"_id">>, 1, Props),
+    Props2 = [{K, V} || {K, V} <- Props1, binary:first(K) =/= $_],
+    #doc{id = DocId, body = {Props2}}.
+
 % Skip patterns
 
 % Build the skip patterns from a config section