Also included it in the tarball

The effect of nonlinearities becomes important at small scales anyway,
so we may as well start from 5 Mpc/h.

After hours of debugging, I finally figured out how to include everything
in the tarball when running `make distcheck`.

As shown here:
https://stackoverflow.com/a/11966104
the key is to use $(srcdir), which denotes the top level directory.

Also, I'm not sure what's the difference between `dist_pkgdata_DATA` and
`EXTRA_DIST`, since both seem to get included anyway (maybe one is an
install target and the other isn't? Idk).

In any case, `make distcheck` should now work, with config files AND
test data included, even though some of the tests will fail since I've
used the CUBA library to generate those, and CUBA is much faster
and more accurate for multidimensional integration (of course,
if the user installs COFFE with CUBA enabled, then the tests will
pass when running `make check`).

Okay, this was very annoying to debug; basically, the flag
`flatsky_density_lensing` turned on just the flat-sky approximation
for density and lensing, and I wanted it to turn on all of the local-nonlocal
terms (so RSD-lensing and d1-lensing as well, though they're zero).
This commit adds that, AND renames the flag into `flatsky_local_nonlocal`.
The results from the COFFE 2 paper remain valid, since the difference
is miniscule.
Also renamed `flatsky_lensing_lensing` into `flatsky_nonlocal`,
since I may implement the other integrated terms at some point
in the future, and added more detailed comments in the settings about
what they actually do.
Finally, left the names `functions_flatsky_[density|lensing]_lensing`
intact for now, so just the flags in `coffe_parameters_t` have been
changed.

"Warning" is now big scary "WARNING"

Default multipoles had double type instead if int

Removed checks for `input_separations` and `input_power_spectrum` since
they make no difference now that COFFE does the check on its own.

It's too long, and unclear; renamed to flatsky_local, because
it turns on the flat-sky approximation for all local terms.

Also, more default separations

Uses defaults now, not sure if it's the best solution,
but it works.

Didn't test compiling without Cuba, so some errors appeared,
should be fixed now.

Provided clearer explanations.
Also removed matter_bias_analytic for now.

`cubacores` used to accept ints, now it accepts pointers
to ints, and the mismatch caused segfaults with the old code.

Also added a `approx_equal_const_epsilon` function,
which uses as defaults COFFE_REL_EPSILON and
COFFE_ABS_EPSILON, respectively, while
`approx_equal` now takes custom values.

Option `only_cross_correlations` controls whether or not
we should only consider the cross-correlations when
computing the estimator/signal.

It's a bit inconsistent to include it in signal.c.
Also, more comments on the various factors

Mostly casting things

Now there's just two, with self-explanatory purposes:
- coffe_integrate_1d
- coffe_integrate_multidimensional
This reduces the need to copy-paste too much stuff everywhere.

signal_integrate_gsl is now called
signal_integrate_multidimensional,
which can be used to integrate any
multidimensional function.

TODO add all of those which we need when
using the analytic method for density-lensing

Extra comma caused compilation errors