$ ~/src/fastText-0.1.0/fasttext cbow -input toy-data.txt -output toy-model -bucket 100
Read 0M words
Number of words:  22
Number of labels: 0
Progress: 100.0%  words/sec/thread: 209  lr: 0.000000  loss: 4.100698 eta: 0h0m -14m

this will make it easier to debug manually

$ ~/src/fastText-0.1.0/fasttext cbow -input toy-data.txt -output toy-model -bucket 100 -dim 5
Read 0M words
Number of words:  22
Number of labels: 0
Progress: 100.0%  words/sec/thread: 199  lr: 0.000000  loss: 0.000000  eta: 0h0m

it cannot pass by design: training is non-deterministic, so conditions
must be tightly controlled to guarantee reproducibility, and that is too
much effort for a unit test

This avoid a null dereference that could previously be reproduced with:

python -c "from gensim.test.test_fasttext;import NativeTrainingContinuationTest as A;A().test_continuation_gensim()"

Move the lower-level FB model loading code to a new module.
Implement alternative, simpler _load_fast_text_format function.
Add unit tests to compare alternative and existing implementation.

This reverts commit 07c84f5.

We have to worry about backwards compatibility if we remove this
attribute, and it's not worth doing that as part of this PR.

Also ran python -m doctest gensim/models/fasttext_bin.py to check the
docstring is correctly executable.

vectors_lockf is only for word2vec.  FastText implementation uses
vectors_ngrams_lockf and vectors_vocab_lockf only.

The order of the words matters.  In the previous implementation, this
was maintained explicitly via the index2word list, but using an
OrderedDict achieves the same thing.

The main idea is that we iterate over the vocab terms in the right order
in the prepare_vocab function.