I think float() should support non-ascii digits but I agree that it would be better to avoid UnicodeErrors and convert them to ValueErrors so that

>>> float('١٢٣٤.٥٦')
1234.56

and

>>> float('½')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: invalid literal for float(): ½

I.e. float should do the C equivalent of:
try:
  s = arg.encode('decimal')
except UnicodeEncodeError:
  raise ValueError('Invalid liter for float(): {}'.format(arg))

Note that int() and Decimal() supports non-ascii chars too.