Celltest always generated for non-empty code cell, and always runs; defaults to %cell if test is empty

[ceball]

Fixes #169, i.e. implements what is described in the opening comment there.

The important changes are in nbcelltets.tests_vendored - the rest is test updates.

Questions:

• Decide what magic comment to use for saying "cell is mocked" (i.e. do not run this cell). Currently # no %cell. Vidar proposed # nbcelltest-skip. Or even possibly go straight to Allow notebook authors to control cell skipping/mocking other than by comments in code? #171? (I'd rather try the current underlying behavior out for a bit, then if we are keeping it, use a better mechanism than comment.)
• Agree it's ok to generate and run a test_cell_3 method (as opposed to not generating/skipping) when no explicit test has been supplied for cell 3, and hence the test defaults to %cell. (This is "case 3 from Proposal for when cells/tests should execute #169.) I think we need some people to try it out in lab etc and give feedback one way or the other.
• In this PR, a code cell that has an empty ast cannot have a test (exception if it does have a test). Does that seem ok? E.g. would it be reasonable to write # note: there is a test at this cell! in your notebook, and have test code?

To do:

• Validate implementation matches Proposal for when cells/tests should execute #169
• Validate it seems to makes sense when using in lab (at least to author!)
• Upstream equivalent nbval fix for bad magics (not how it's done here) filed an issue/started a PR/it's not related to this PR

Also:

• Default text in lab's CELLTESTS changed (more instruction, reflects default)
• Distinguish "only whitespace" from "empty ast" more clearly (previously, "is empty" was "empty ast"). This PR does not change the fact that "empty cell" is "empty ast".
• Closes Why does running skipped tests during nbcelltests' own test run terminate pytest? #117 (no skips now anyway)
• Added tests showing Document (and ideally clean up) rules around using %cell #173 (known fail for now)

[ceball]

@vidartf I have requested a review even though there's a not-yet-done to-do item (that just relates to nbval's handling of bad magics). Meanwhile the other changes and questions are ready for review.

[ceball]

@vidartf I have made this not be WIP - I have removed the nbval change.

[vidartf]

I didn't fine-comb it, but it LGTM!

[ceball]

We might revisit some things here, but I agree we should merge.