Handle a TODO in the parser when the previous span starts after the current one

[asterite]

This can happen if the previous token location is in a different file. In that case it probably makes sense to not merge the location, but this issue is captured to avoid growing the original PR.