Improve ufloat_fromstr documentation and parsing

[Cs137]

The ufloat_fromstr documentation does not mention how the function generates an uncertainty if the string representation does not contain one. Is it always 1 to the last significant digit? Also, I doubt that it makes sense to assign an uncertainty to a NaN value.

Examples:

• ufloat_fromstr("123") -> 123.0+/-1.0
• ufloat_fromstr("987") -> 987.0+/-1.0
• ufloat_fromstr("0.9999") -> 0.9999+/-0.0001
• ufloat_fromstr("nan") -> nan+/-1.0

[newville]

@Cs137 Yes, it does default to an uncertainty of 1 in the least significant digit. That seems like an OK behavior compared to the otherwise obvious default of "raise ValueError".

I would say the same about what to do with the string "nan": that result seems acceptable, though "raise ValueError" might be a better choice.

Adding a note and/or example about this in the docs as part of #285 would be fine.

[Cs137]

Thanks for the answer @newville, I will pick up your proposal and add a note about it in #286.