fix: sanitizer double-escaped entities ( /&); dot-prefix volumes dir

- chat.SanitizeHTML/SanitizePlain decoded-then-escape text runs, so HTML
  entities the browser emits (e.g.   for the space after a mention chip,
  or & for a typed "&") no longer render as literal " "/"&".
  Re-escaping keeps output XSS-safe. Adds regression tests.
- Rename ./volumes -> ./.volumes so `go build/vet/test ./...` skip the mongo
  data dir (700-perm files); Makefile lint now gofmt's tracked files only.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

internal/chat/sanitize.go

@@ -31,10 +31,10 @@ func SanitizeHTML(in string) string {
 		if c != '<' {
 			j := strings.IndexByte(in[i:], '<')
 			if j < 0 {
-				out.WriteString(html.EscapeString(in[i:]))
+				out.WriteString(escapeText(in[i:]))
 				break
 			}
-			out.WriteString(html.EscapeString(in[i : i+j]))
+			out.WriteString(escapeText(in[i : i+j]))
 			i += j
 			continue
 		}
@@ -118,6 +118,15 @@ func SanitizeHTML(in string) string {
 	return out.String()
 }
 
+// escapeText normalizes a text run from (possibly entity-encoded) input HTML:
+// decode existing entities, then re-escape. This keeps a single round of
+// escaping so entities the browser emits — e.g. &nbsp; for spaces around inline
+// elements, or &amp; for a typed "&" — don't get double-escaped into visible
+// "&nbsp;"/"&amp;" text. Re-escaping keeps the output XSS-safe.
+func escapeText(s string) string {
+	return html.EscapeString(html.UnescapeString(s))
+}
+
 // splitTag separates the tag name from its raw attribute string.
 func splitTag(raw string) (string, string) {
 	for i := 0; i < len(raw); i++ {
@@ -219,10 +228,10 @@ func SanitizePlain(in string) string {
 		if in[i] != '<' {
 			j := strings.IndexByte(in[i:], '<')
 			if j < 0 {
-				out.WriteString(html.EscapeString(in[i:]))
+				out.WriteString(escapeText(in[i:]))
 				break
 			}
-			out.WriteString(html.EscapeString(in[i : i+j]))
+			out.WriteString(escapeText(in[i : i+j]))
 			i += j
 			continue
 		}