Replace hardcoded pass sequence with a scheduler

[jdpage]

Passes are now tagged with information about what they do, and are automatically placed into the correct order. The scheduling algorithm is designed such that the same input will always produce the same output, so that the order that passes tests is the same order used in production. This can be disabled for testing purposes.

[coveralls]

Pull Request Test Coverage Report for Build 439

• 114 of 139 (82.01%) changed or added relevant lines in 11 files are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage decreased (-0.8%) to 87.202%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/jeff65/ast.py	3	5	60.0%
src/jeff65/pattern.py	11	13	84.62%
src/jeff65/graph.py	44	54	81.48%
src/jeff65/init.py	1	12	8.33%

Files with Coverage Reduction	New Missed Lines	%
src/jeff65/init.py	1	22.09%

Totals
Change from base Build 431:	-0.8%
Covered Lines:	1630
Relevant Lines:	1847

---

💛 - Coveralls

[jdpage]

@woodrowbarlow Assigned this to you not because I necessarily want you to go over the code--though feel free to--but because I'm not sure if this is a clever idea or a dumb idea and I want your input.

[woodrowbarlow]

so, without looking at the code, my initial thoughts:
a hardcoded pass sequence gives us 100% control and flexibility. do you believe the code is at a level of complexity where manually managing a hardcoded pass sequence would be more painful than (worst case scenario) having to modify the scheduler or fight with it to get the ordering the way we want?
having the pass sequencing be algorithmicly determined based on metadata seems like a good idea eventually. do you think we're there?