[MERGE] Move the notes on writing tests out of HACKING into a new file, and improve them.

[Adrian Wilkins]

Robert Collins wrote:
> As far as getting everything passing again being a blocker to having PQM
> lock-step test, I don't think it is - we can decorate the current set of
> failures fairly easily and new commits would only fail on new problems -
> letting the win32 interested parties whittle the test suite down.
> 
> -Rob

This was part of my reasoning for suggesting more structured output from 
the test suite ; to put it in bzr-esque terms, you could have a test 
inventory, with kinds of "passed", "failed", "errored", "added", 
"removed", "skipped", etc, and pass revisions that produced a 
non-negative delta (e.g. no new failures) for merging. In addition it 
becomes much easier to have pretty web pages showing where win32 hackers 
can concentrate their efforts. Decoration of the code not required 
(although this is certainly one way to do it).