fix: improve embeddings ranking with full query context

The embeddings ranking by symbols is now performed using the complete query
context including chat history, which provides more accurate results when
finding relevant code sections. Also improves references check in chat UI
using utils.empty helper.

Author: deathbeam

lua/CopilotChat/context.lua

@@ -656,13 +656,6 @@ function M.filter_embeddings(prompt, model, headless, embeddings)
 
   notify.publish(notify.STATUS, 'Ranking embeddings')
 
-  -- Rank embeddings by symbols
-  embeddings = data_ranked_by_symbols(prompt, embeddings, MIN_SYMBOL_SIMILARITY)
-  log.debug('Ranked data:', #embeddings)
-  for i, item in ipairs(embeddings) do
-    log.debug(string.format('%s: %s - %s', i, item.score, item.filename))
-  end
-
   -- Build query from history and prompt
   local query = ''
   if not headless then
@@ -674,6 +667,13 @@ function M.filter_embeddings(prompt, model, headless, embeddings)
   end
   query = query .. '\n' .. prompt
 
+  -- Rank embeddings by symbols
+  embeddings = data_ranked_by_symbols(query, embeddings, MIN_SYMBOL_SIMILARITY)
+  log.debug('Ranked data:', #embeddings)
+  for i, item in ipairs(embeddings) do
+    log.debug(string.format('%s: %s - %s', i, item.score, item.filename))
+  end
+
   -- Embed the query
   table.insert(embeddings, {
     content = query,

lua/CopilotChat/ui/chat.lua

@@ -252,7 +252,7 @@ function Chat:render()
 
     self:show_help(msg, last_section.start_line - last_section.end_line - 1)
 
-    if self.references and #self.references > 0 then
+    if not utils.empty(self.references) then
       msg = 'References:\n'
       for _, ref in ipairs(self.references) do
         msg = msg .. '  ' .. ref.name .. '\n'