Fill offset. Fix some linenums and columns with 0 value.

[juanjux]

This adds the offset to the Python nodes, using the transformation parser. It also fixes bug #35 that was a requirement for the former to work. This fixes 2 of the 3 tasks required for fixing #30.

For the next PR I will fork pydetector into bblfsh as @abeaumont suggested some time ago because it will require extensive changes.

PR includes:

• Use a TransformationParser with a FillOffset function to add the offset.
• Fix bug with some columns showing at 0.
• Fill lines for nodes that the Python's native AST doesn't fill.
• Update integration tests with the right offsets.
• Added new integration tests to avoid regressions of Node positions not being according with specs #30 and Parsing an empty file produces a fatal error #25

[abeaumont]

Function name doesn't match with real one. I'd suggest to remove the function name from the documentation, since it's redundant and error-prone

[juanjux]

Removed.

[abeaumont]

Is this status change expected? The same question applies to the next UAST files

[juanjux]

Oh f*ck I was so centered on the non zero column number and offset that skipped these ones, good catch, thanks.

[juanjux]

Updated. I added line numbers to some nodes that didn't have them in the native AST (which set them at 0 in the UAST). This fixes all of the previous errors except for the new empty.* test (expected, since it's empty, it's tests that the Python driver doesn't crash on this case like happened with a previous solved issue) and the string_fstring.*.

This last one does the curious thing of setting the line numbers the code inside the fstring's {...} as starting from line 1. Shouldn't matter much because the next PR should check and fix all those "questionable" locations of the native AST.