* Fixed a bug in the coarse grid density correction that lead to a shift in the correction in some rare cases
* Fixed a bug in restricting fine grid noise when the double size fine grid (for FFT) is actually larger than the domain.
* Added new log file routines that use printf syntax
* Added log file output to most parts of the code
* Removed the stats file output, it is now appended to the log file
* Fixed a normalisation bug in the hybrid Poisson solver
* Added calculation of the exact T(r=0) by summing modes rather than by analytical integration
* Fixed a bug in the deconvolution that caused the Nyquist mode to be excluded in the deconv.
* Fixed a bug in the random number generator when using single thereaded FFTW.
New conservative transfer function scheme.
New random number generator that preserves coarse Fourier modes instead of applying Hoffman-Ribak
Some minor bugs fixed.
* Moved hybrid poisson solver from convolution_kernel.cc to poisson.cc
* Fine-tuned deconvolution schemes for various running modes
* Added lots of doxygen documentation
improves the solution.
Also fixed another bug related to the previous grid placement correction with align_top,
size corrected grids have to be enforced to have an even offset.
LLA baryon density might deviate from the mean density by up to a few percent.
This is now corrected explcitly by subtracting the mean after calculation
Grid sizes have to be a multiple of 4 during density convolution when using
isolated boundary conditions. Otherwise the padding region is not coextensive
with coarse grid cells. This is now enforced during density generation but
is readjusted afterwards.
Note to myself: don't fix signs while listening to talks!!
Other mods/fixes:
* 2LPT is now supported also with distinct baryon/DM
power spectra. This creates now an overhead when
using Eisenstein&Hu or BBKS. But who cares.
* Reverted to determinant formulation of LLA rather than
second order expansion. This is much more accurate.
* Added baryon density filter. Baryon density field is now
consistent with CIC DM density. This is achieved by
convolving with an additional grid cell filter (and
no deconvolution).
* Tested LLA, skewness is now exact to a few per cent
but tends to be systematically higher.
* 2LPT needs a bit less memory now in multigrid mode
(FFT based 2LPT still needs humongous amounts).
* Cleaned up code output.
