Add high-level functional tests

[glesica]

We should check for the following:

• Doesn't crash on known good inputs
• Handles known bad inputs reasonably
• Produces valid (if not necessarily correct) trees on known good inputs
• Modifies behavior correctly in the presence of different flags

Additionally:

Test harness and an automated, high-level test suite, compare against RapidNJ and QuickJoin, use tree distance as a heuristic
Generate some random sequences and run them through each tool, diff them against one another

Also run some very simple sequences (like the examples, perhaps) to guard against regressions and to keep the documentation correct (gold file tests) [TW suggested using random alignment generator, but with fixed seed]

Test cases (minimum): shorter than the SSE registers, exactly equal to, and larger than the registers (length should be a prime number)