!!!Speller conversion test bench

Done:

* augmented makefile to produce small test speller transducer, based on list of
  test words

Next steps:

* check the size of the PLX output - if reasonable, store it in svn as reference
* add spell checking of the test words as well, store the speller output in svn
* edit manually the speller output to reflect what we want
* edit manually the PLX codes until we get what we want from the speller
* change the regular PLX conversion to reflect the manually tuned test PLX codes

!!Problem

This will still be one large test file. What we really want is a number of very small and focused tests.

!!Possible solution

* define the test cases in terms of the words (ie classes) involved
* split the PLX data on the test words, so that only data relating to the words for each test case is included.

The test cases should be stored like this:

{{{
$GTHOME/gt/$GTLANG/polderland/testing/caseN/words.txt        # in svn
                                            expected.plx     # in svn
                                            expected.speller # in svn

                                            speller.lex # <- binary speller file
                                            actual.plx       # NOT in svn
                                            actual.speller   # NOT in svn

                                     /caseAll/ concat all words, sort
                                            expected plx
                                            expected speller
}}}

Output from each test case:

{{{
FAIL
PASS
}}}