Ignore MKL-based sparse solvers when -Dwith_mkl is set to false but the
specific solvers are enabled.

Add basic support for GENCODE and miRBase FASTA format.

Add support for reading RSEM, HTSeq and TSV counts.

Use a geometric mean when replicated gene expression are provided with a
log(X+1) transformation.

The quantification is based on the assumption that the total
concentration of protein-coding transcripts and microRNA in the
cytoplasm is preserved across tissue types. We use a model of a HeLa S3
cell to estimate the number of species and the cytoplasmic volume.

It's already implied by -Dwith_mkl=true when testing the LAPACK solver.

Only check for the subset of positions of a target where a seed can fit
and let compute_score perform the remaining checks.

Remove code for detecting FASTA formats and instead rely on the user
providing explicitly the format via a specialized flag.

Add --ncbi-targets, --gencode-targets, --mirbase-targets flags to the
binary.

Use /opt/intel/mkl as a default path for mkl_root.

Each warning that suggest a dependency now also include a corresponding
configuration flag.

We already have explicit flags for enabling MKL LAPACK and FFTW3, so the
warning messages have been removed.

This strategy prevents many useless allocations for lines in-between
sequence records by positioning the stream on the beginning of a new
record.

It also simplifies the parsing code since we don't need to check line[0]
anymore.