The codec code has a few (performance) issues:

 * nonspacing_diacritical_marks should be a set for fast lookup
 * ord(c) in range(0x00, 0xA0) should be rewritten using < and >=
 * result += bytes([ord(c)]) has exponential timing (it copies
   the whole bytes string for every single operation); better
   use a bytearray and convert this to bytes in one final step
 * the error messages should include more useful information
   about the cause and location of the error, instead of just
   UnicodeError("Unacceptable unicode character") and
   raise KeyError

Please also check whether it's not possible to reuse the charmap codec
functions we have. Thanks.