Issue Description:

Sometimes a Unicode non-breaking space character (U+00A0) slips into a source file, which can lead to inscrutable errors like this:

<stdin>:1:7: error: '=' must have consistent whitespace on both sides
var x = 1
<stdin>:1:8: error: invalid character in source file
var x = 1

(Note that on the command line we can't even highlight the non-breaking space, because we don't know its display width. That's a separate problem.)

It would be nice™ if non-breaking spaces were recognized by the lexer and treated as spaces, with a dedicated warning and fix-it: "non-breaking space (U+00A0) used instead of regular space".