feat: validate ?columns mutation targets based on schema cache (#2542)

This returns an error for trying to update or insert into invalid columns, without hitting the database. This change also switches from `json_populate_recordset` for these operations `json_to_recordset` which should make no functional difference except allowing future flexibility.

src/PostgREST/Error.hs

@@ -81,6 +81,7 @@ instance PgrstError ApiRequestError where
   status UnacceptableSchema{}    = HTTP.status406
   status UnsupportedMethod{}     = HTTP.status405
   status LimitNoOrderError       = HTTP.status400
+  status ColumnNotFound{}        = HTTP.status400
 
   headers _ = [MediaType.toContentType MTApplicationJSON]
 
@@ -216,6 +217,11 @@ instance JSON.ToJSON ApiRequestError where
     "message" .= ("Could not choose the best candidate function between: " <> T.intercalate ", " [pdSchema p <> "." <> pdName p <> "(" <> T.intercalate ", " [ppName a <> " => " <> ppType a | a <- pdParams p] <> ")" | p <- procs]),
     "details" .= JSON.Null,
     "hint"    .= ("Try renaming the parameters or the function itself in the database so function overloading can be resolved" :: Text)]
+  toJSON (ColumnNotFound relName colName) = JSON.object [
+    "code"    .= ApiRequestErrorCode18,
+    "message" .= ("Column '" <> colName <> "' of relation '" <> relName <> "' does not exist" :: Text),
+    "details" .= JSON.Null,
+    "hint"    .= JSON.Null]
 
 -- |
 -- If no relationship is found then: