For ACLiC, we have to generate ".out" files which contains dependency
library name separated by " ". This was done by creating
std::vector<(identifier from rootmap file), (library name)> and comparing this with decls from the input file.
However with modules, we want not to use rootmap files. Module's name
already contains information about library (modulename should be
consistent to the library) so we need not to store all information in
the vector anymore.

optimize SGD and ADAM update calculations. For ADAM define specific update functions which are re-implemented for each architecture

It is not used in TVirtualX interfaces.
It is re-declared in TGLUtil.h include.
Was introduced 14 years ago with the commit:

https://github.com/root-project/root/commit/fc7ab2b

In my mind - it is wrong place

We are flooded by warnings of shadowing enum constants, e.g.

include/TDictionary.h:189:7: warning: declaration shadows a variable in the global namespace [-Wshadow]
      kNone              = ROOT::kNotSTL,
      ^
include/GuiTypes.h:87:16: note: previous declaration is here
const Handle_t kNone = 0;
               ^

We will need to address them, but to keep the warning / email volume reasonable, disable for now.

We need to make sure the ROOT C++ kernel of Jupyter runs with the
same Python version that ROOT was built with.

Since running the kernel with a particular Python installation
cannot be enforced by the user by setting the $PATH variable
(Jupyter ignores that variable and picks Python from the
system), we need to enforce it from the kernel file.

We can't define a full path to the right Python binary in the
kernel.json file because that path could be different in the machine
where ROOT is built and the machine where the C++ kernel runs.
The best we can do is include the major and minor version in
the kernel.json file, so that we make sure the kernel is not
executed with an incompatible version, which would likely lead
to a crash.

in case no model is provided, this commit triggers the automatic creation
of names and titles of graphs and histos based on the input column names.

In addition to the code changes, the existing documentation has been
improved.

This reverts commit 8ac9554e.

This reverts commit 7cf40ecd.

Move several helper classes into sources

which is its dautgher, in order to restore the symmetry with the C++ interfaces,
where TDirectory::Get does not exist.

this allows to create files with this factory pattern

myFile = ROOT.TFile.Open("myfile.root")

the pythonizations are three:
1. A C++ pyz to provide attr syntax. For example:

myfile.mydir.mysubdir.myHist.Draw()

2. A C++ pyz to allow users to invoke the templated method TDirectory::WriteObject<T>. For example:

h = ROOT.TH1F("h","",10, 0, 1)
myfile.WriteObject(h)

3. a Python pyz which implements TDirectory::Get in terms of the pythonized attr syntax

these are internal Cppyy utilities. In principle it would be better to
expose them in cppyy and use them in PyROOT and not duplicate the code.
This commit at least avoid to duplicate the code only once.

pythonisations are lazy: upon class usage, the class gets pythonised, not before.
For some core classes though, for example TDirectory, lazy pythonisation is not possible.
This change introduces non-lazy pythonisations to address such cases.
The syntax for lazy pythonisations becomes

@pythonization()
def pythonize_myclass(klass, name):
  ...

to make a pythonisation non-lazy:

@pythonization(lazy=False)
def pythonize_myclass(klass, name):
  ...

__attribute__ is not supported on Windows, but it is internally needed by Cling for autoloading and Clad rely on __attribute__((annotate("D")))