As a followup to last revamp:

 *  Use more extensible test definition format,
 *  separate testing framework from setup script and split to includes,
 *  update test scripts as well as data,
 *  consolidate test output,
 *  and make some small improvements/fixes on the way.

 *  Enable running all tests via make,
 *  simplify tests (lay off data/tmpdir manipulation to framework),
 *  set rules for naming, exit statuses, etc,
 *  remove beakerlib dependency.

Use the meta-library and handle that oputside __echo (i.e. do not try to
import it kajillion times).

Purpose of this library is to import the actual pretty-printing library
based on FFOO_MKPRETTY value or fall back to mkpretty_plain.

A library can now have init function to allow for more dynamic behavior.
The function's name must be __ffoo_NAME__init, where NAME is the library
name.

Note that the init exit status is used as exit status of the `ffoo
import NAME` call (or `ffoo try_import NAME`) and overrides exit
status of the sourcing (which must always be zero anyway).

try_import does not aim to import anything, whereas this function's aim
is to import stuff.

Code include expansion overwrote result of variable name expansion.

(This had no effect so far since there are no variables in markdown
files.)

Move expansions to functions and comment them properly there.

Die by default but allow "just try"

 *  Split into functions as appropriate;
 *  add docstrings;
 *  reduce redundancy;
 *  run everything in C locale (=> update oracles),
 *  add subtest boundaries to output.

Instead of creating SemVer name on-premise, it will be hard-coded inside
the tarball

Now we can keep examples as a "real" shell scripts, taking advantage
of ability to test them or edit them with proper syntax HL.

Setup will now automatically insert them and indent properly as code.

Small drawback is that we cannot build manpages separately, but since
all building is to be trivial this is not a big deal.

...to better demonstratr new merging capabilities

Strict mode now preserves also comments, which makes sense when iniread is used only
to merge ini files verbatim.

It's not specifed yet which is to be picked, but as FFOO_INI_PATH is now
respected properly, it's more intuitive to pick the last--the last item
on FFOO_INI_PATH wins, which is similar to how PATH works.

Using empty element results in checking "/file.ini", which is not what one
would want.  (One can always explicitly provide path as argument.)

Increase readability: decrease complexity of the build function and
preserve wandering off too much against 80-char "buoy"