[RF][Docs] Documentation updates / fixes collected througout june.

Conda sets a prefix for cross-compilation. That prevents CMake from finding headers of
the current ROOT build, see https://sft.its.cern.ch/jira/browse/ROOT-10210 and
https://github.com/conda-forge/root-feedstock/issues/38 . Instead, turn off CMAKE_FIND_ROOT_PATH
for headers searched through relative directories, or absolute ones that start with the
build or source directory. (Do not assume that CURRENT build dir is a subdir of CMAKE_BUILD_DIR.).

- Add a tutorial for the Barlow-Beeston method, rf709_BarlowBeeston.C.
- Remove legacy iterators from RooHistConstraint and RooParamHistFunc,
which are used to implement the Barlow-Beeston method.
- Replace iterative logGamma in RooHistConstraint by std::lgamma
- Make sure that bins with zero entries don't get touched by MINUIT.
This reduces the number of degenerate parameters without any effect, and
stabilises the covariance matrix.
- Update documentation of the respective classes.

* Allow to display tutorial images with JavaScript

* use find instead of find_first_of

* Output first <div>  then <script>

* Make sure each canvas size is the right one in case of several canvases.

When integrating a region next to a narrow Crystal Ball shape, the integral
can evaluate to zero. Since RooFit cannot deal with zero integrals,
the CBShape will now return the smallest possible double.

See also https://root-forum.cern.ch/t/fit-a-double-crystal-ball-with-roofitranges/34599

SPlot needs to run a fit to compute sWeights. In doing that, it ignores ranges and other fit options that the user might have been using.
- This commit enables passing additional RooCmdArgs to SPlot to get ranges
and integrals correct.
- Update & format documentation accordingly.

See also https://root-forum.cern.ch/t/splot-does-not-work-with-fixcoefrange/34701

The package name in the call to find_package_handle_standard_args()
function must match what is used in the call to find_package() and
the find module name (e.g. FindUnuran.cmake). If there is a mismatch,
then find_package(Unuran REQUIRED) will not cause the desired failure
during configuration when the package is not found. CMake will define
both Unuran_FOUND and UNURAN_FOUND for compatibility reasons.

This option is ported from ROOT 5.x but disabling it makes
compilation break since a long time. Fixes: ROOT-9449.

This fixes the warning: ignoring return value of ‘char* fgets(char*, int, FILE*)’, declared with attribute warn_unused_result [-Wunused-result]

This fixes a bug visible only in runtime_cxxmodules where the include guards
are ignored. This can happen when we are building a two modules which need
the same definition. In this case the usual include guards will only work if
we call a proper `#include "something"` and this something should be defined
in a modulemap.

There is no better solution for this at the moment. This is a rare case which
will likely not affect external use-cases.

Revert "[cxxmodules] Fix R.M by displacing TEmulatedTuple only without cxxmodules"

This reverts commit a74297ac and puts in place
a better fix. Now roottest-root-io-tuple-make should work with modules.

Patch by Philippe Canal and me!

This patch makes it more difficult for people to avoid the 'automatic' memory
ownership done by TMemFile. It forces people to explicitly construct the
data structure describing memory blob and makes it easier to grep for.

ROOT uses out of source builds only, so we do want to see when
files are created within the source directory.

Use FindPackageHandleStandardArgs.cmake to handle common options
passed to find_package(), like QUIET, REQUIRED, etc.

When builtin_vdt=ON, vdt headers are in ${CMAKE_BINARY_DIR}/include
and libvdt.so is in ${CMAKE_BINARY_DIR}/lib, so we don't need to add
the extra include directory. Once ROOT is installed, the Vdt headers
and library will be along with ROOT ones, so they can also be found
without needing anything else either.

Fixes: ROOT-10197

I.e. the Streamer no longer override the user choice