Fixes to noecho parsing

[hugetim]

Related to #16

Known issues remaining:
quietly and capture blocks should not prepend noisily to commands within the block

(I'm starting to think basing the noecho code on program blocks (#15) would have been easier than the quietly-noisily approach, after all.)

[hugetim]

Tracking when a Stata command is inside a quietly or capture block is more complex than tracking in_program because {} can be nested, whereas the different types of programs (program, mata, python) cannot. So doing it right requires matching open and close brackets.

That is why, at this point, I suspect reviving #15 with the benefit of the new helpers for addressing #delimit and program definitions (separating them out and running them separately) may end up simpler. But we may continue bumping up against other Stata features, in either case.

[ticoneva]

I agree it's probably a good idea to use the PR15 method unless a #delimit or program define (and its abbreivation) is detected.

[hugetim]

For future reference, here's a notebook with the manual tests this PR was trying to address (with the results shown for the latest 'master' commit, 0898830, with configuration echo = None).
pystata-kernel manual tests.zip