Improve support for multiple blame ranges

[holodorum]

This PR implements various improvements suggested in PR #1973.

The main change is converting the BlameRanges struct, into an enum to support both OneBasedInclusive and ZeroBasedExclusive range formats. The FullFile variant, denotes that the entire file is to be blamed, serves as a more explicit substitute for a previously used "empty" range.

[Byron]

Thanks a lot for the follow-up!

@cruessler would probably be the one to do the first review, and I will get it merged right after.
Thanks everyone

[cruessler]

Thanks a lot for the follow-up!

Before I start reviewing this PR in detail, I have a few questions regarding the high-level API. In particular, it seems as if the decision to change to_zero_based_exclusive(&self, max_lines: u32) -> Result<Vec<Range<u32>>, Error> to to_zero_based_exclusive(&self, max_lines: u32) -> Result<BlameRanges, Error> brings a couple of downsides with it. My main concern is that it leaks implementation details: the caller now has to know about BlameRanges’ internals while that was previously not the case. Also, two of BlameRanges’ methods are now fallible, forcing the caller to deal with errors that are coupled to the struct’s internals.

I think the proposed API can be simplified and stay more insulated from calling code by reducing the number of enum variants to WholeFile and PartialFile (or something more appropriately named) and staying with to_zero_based_exclusive(&self, max_lines: u32) -> Result<Vec<Range<u32>>, Error>. We could even go so far as to stay with the existing design that treats an empty ranges as covering the whole file.

What do you think?