fix bug where scm_from_utf8_stringn would not detect bad utf-8

fix bug where scm_from_utf8_stringn would not detect bad utf-8

* libguile/strings.c (scm_from_utf8_stringn):
* libguile/symbols.c (utf8_string_equals_wide_string): The "bad UTF8"
  return from u8_mbtouc is a 0xfffd character, not a negative byte
  length.  Fixes a bug in which invalid UTF-8 would not be caught.

* libguile/bytevectors.c (scm_utf8_to_string): Use scm_from_utf8_stringn
  directly.  Just a little cleanup.

* test-suite/tests/iconv.test ("narrow non-ascii string"): Add test for
  parsing bad utf-8 with substitution.