be more careful about final LUs in analysis mode (#104)

[mr-martian]

I think this fixes #104 and tests pass, but I'm not totally certain whether any weird things might happen due to, say, stepping by null.

[unhammer]

There are tests for NUL flushing, so it's probably ok, but perhaps try checking that it works with the rest of the pipeline by running a local apy instance? (put your version of lt-proc in PATH, apt install apertium-apy apertium-spa-cat and curl "http://localhost:2737/translate?markUnknown=yes&langpair=spa|cat&q=sometest")

[TinoDidriksen]

From testing a larger corpus, it seems to work. There's only extra words in the output - no changed ones.