(PUP-2411) Make Program/Locator contain byte vs char offset encoding

As discussed in the ticket. Ruby before 1.8.7 has no way to efficiently
scan a string by character (it is done per byte). In Ruby 2.0.0 the
StringScanner can respond with character positions instead of byte
positions. It is however slower than using the byte positions.

To make 1.8.7 or 1.9.3 encode strings in chars would be prohibitively
slow (a string has to be constructed to be able to know its length for
every token). In Ruby 2.0.0 it is more or less a wash (they byte way of
calculating this is however still faster (when last measured).

This fix makes it possible to record the information in the model if the
offsets in the model are byte or char based. Currently, there is no
implementation that uses char based, so this is preparation for the
future and making the AST model's API more stable.

As a comparison on other platforms (JVM) it is far easier and more
efficient to get the char positions than the byte positions).