Hey, could windows labels (-l) be used for another thing ?

This test is to ensure that the sequences are UNSEG. The locus should not be tested here.

The test for pseudo-germline MAX12 is useless since we have the e-value.

When the sequence is not segmented, the total number of k-mers is used.

Previously, many TOO_FEW_ZERO situations were reported as TOO_FEW_V.

Fix 6ceffe02. We forgot to re-compute the total e-value.

We thus get evalues.

And correctly initialize 'because' in segment.cpp

Now the stats on each locus are the number of segmented reads with windows.
This is more coherent with the rest of the stats (in particular the reads/clones ratio).

To be too short for the window is just a non-segmentation/non-analysis cause.

Previously, the default -w was 40 for VJ, and 60 for V(D)J, except in
multi-locus identification (-g), when it was 40 even for V(D)J recombinations.
For the majority of cases, setting -w to 50 is thus more conservative.
Many users already tested -w 50 (or the less conservative -w 40) for IGH and
found correct results.

Now we have a uniform window size for all locus, allowing a better indexing
and comparison of windows across experiments and configurations.

Light refactor.
The e-value is actually computed deep into each KmerSegmenter, so it is cleaner to store e-values at this step.
The MutliKmerSegmenter task is only to give the good multiplier.

- stanford*: one sequence in S22 is not segmented
- trd-dd2-dd3*: the short Dd2-Dd3 example has a evalue slightly above 1.0, we set -e 10
- chimera:  one failed test is now passing thanks to e-value

The e-value threshold just works. It is thus enabled by default in the next release.

Thank you Valgrind, again.

Thank you Valgrind.

Get an approximation of the number of sequences in a fasta file
by interpolating from the size of the first 200 sequences.

Note that input->tellg() was not working (not compatible with igzstream ?)