Multiplying by a factor of 1/1000.f (rather than dividing by 1000.f) directly
introduces an error of 1 ULP. With this patch, an exact comparison of a
floating point literal with the deserialized F1000 form representing it is now
guaranteed to be successful.
In addition, the maxmium and minimum safely representible floating point
numbers are now well-defined as constants.
Use wide_to_utf8 and utf8_to_wide instead of wide_to_narrow and narrow_to_wide at almost all places.
Only exceptions: test functions for narrow conversion, and chat, which is done in a separate commit.