Tests for m4 comparison

[thesamesam]

The m4 comparison tools are the most complex and a lot of the time taken when changing things is in making sure we don't break cases.

A simple test suite, maybe just with diff on output, to make sure we pick up expected cases would help a lot.

[hlein]

Good call, I think we'd need, say:

• knowngoods: ~2 git repos with multiple revisions of .m4s, including serial changes, missing serial, etc.
• test projects: ~2 project trees with some .m4s that match, some which match when comment-stripped, some new, some serial increment, some serial jump, some identical across the test projects...

To avoid sticking binary goo in the repo (for obvious reasons!) do we instead need a make corpus target that builds the desired git repos locally on the fly prior to make test?

Ties in the future as well to:

• fuzzy hashing to pick the best diff candidate rather than "first with a serial match"
• pathological filenames, which we still need to catch
• grouping of similar-unknowns across multiple projects