* Build-Depends: Drop versioned constraint on libgsl-dev.

Changes-By: deb-scrub-obsolete

growth_function.c: Introduce Kasai (2010 arXiv:1012.2671) Eq (5) to
speed up non-EdS flat FLRW growth factor calculations by factors of
about 4, 3, 10, for the zeroth, first, second derivatives,
respectively.

growth_function.c: Introduce missing return line for EdS case for
second derivative of the growth function - this should speed up the
EdS case a little. The old calculation was OK, but unnecessarily
did a lot more work than needed.

FLRW_background.c: Add flat FLRW (non-EdS) second derivative of scale factor,
a_ddot_flat_FLRW.

make check: Extend test_growth_function.c, test_scale_factor.c
for checking second derivatives between EdS and nearly-EdS flat FLRW.

Benchmarks have been added to most of the lib/test_*.c functions.
Some options need turning on with #define's since they are only
for manual testing. The non-EdS growth function is what slows
down LCDM calculations versus EdS by a factor of about 20-30.
A suggestion to try Kasai's (2010, arXiv:1012.2671) equation (5)
has been added to the TODO list, although it might not speed
up the calculation by much. All 7 default test functions and
the 3 developer test functions are successful in this commit.

This restores the regression test `test_biscale_partition` back to
its original settings since the last few fixes have removed the
bug; this commit reverts commit 295f133d.

Some constants for gsl_monte_vegas_integrate were defined without
explicit typing. This commit casts `(size_t)` explicitly to reduce the chance
of typing mismatches.

See the TODO note "Add some mechanism for checking ...". That TODO is
high priority, but pending that, the check for already calculated
power spectrum normalisation background_cosm_params.correct_Pk_norm_known
can be checked against 1 rather than just being non-zero. This gives
a chance of about 2^{-32} of randomly getting 1, which is low enough
for the short term.

See e.g. https://stackoverflow.com/questions/9225567/how-to-print-a-int64-t-type-in-c
https://stackoverflow.com/questions/2524611/how-can-one-print-a-size-t-variable-portably-using-the-printf-family
for recommendations (C99) on printf formats for 64-bit unsigned integers
and size_t, to avoid having to switch between %lu and %llu depending on which
machine/compiler default is chosen. Affects
lib/Omega_D_precalc.c, lib/alloc_big_array.c, lib/sigma_square_invariant_III.c .

On a Knoppix x86_64 intel system with gcc 8.2.0 the numerically
Newtonian phases of two of the test cases do not last as long as they
do on Debian stretch with gcc 6.3.0 . This could either be to
improved accuracy or worsened accuracy, but it's an issue of
accuracy, not a software pipeline issue. This should not block
basic installation, but is worth investigating (TODO).

Standards are important, but this change is only a claim that the
package is compliant with the updated standard; it doesn't change
anything in the code or build structure themselves. So I'm not
proposing a new debian hypened version and build - I'm just doing
updating to salsa git now so that it's in the pipeline for when a more
substantial change is needed.

The previous commit used the `isnormal()` macro, which fails on
zero or subnormal numbers. This commit replaces isnormal by
`isfinite`, since zeros and subnormals are OK, especially when
checking for reproducibility of calculations in unit tests.

Numerical tests for some calculated value to be close to an
expected value can give a clean pass if the calculated offset is
a NaN, since (fabs(NaN) > 1e-3) is false (e.g.
https://www.gnu.org/software/libc/manual/html_node/Infinity-and-NaN.html
`x == x is false if the value of x is NaN` ), so
isnormal(calculated_error) needs to be checked too.

modified:   lib/test_alloc_big_array.c
modified:   lib/test_biscale_partition.c
modified:   lib/test_growth_function.c
modified:   lib/test_gsl_monte_carlo.c
modified:   lib/test_ode.c
modified:   lib/test_power_spectrum_flatspace.c
modified:   lib/test_scale_factor.c

modified:   lib/Omega_D_precalc.c
modified:   lib/alloc_big_array.c
modified:   lib/backward_compatibility.c
modified:   lib/sigma_square_invariant_III.c