Keep knowledge of encoding for text properties around

[mvz]

Working on #14 brought to light that reading text properties results in a different type of result depending on whether the property type is PT_STRING8 or PT_UNICODE. In the latter case, the result is a Perl string (a sequence of Unicode code points), while in the former case, it is a sequence of bytes.

After reading the property, the knowledge of how it was encoded is discarded, so subsequent code needs to either guess at the data type of the property value, or just ignore it.

It would be better to keep the full knowledge of underlying data and type until the property is used. How it is to be decoded in the case of PT_STRING8 depends both on which property it is, and on the value of the PidTagInternetCodepage and PidTagMessageCodepage properties.