Often, when designing an analysis, it happens to deal with empty RVec instances,
for example after a selection cut. This change allows in an ergonomic way to
deal with this case without imposing overhead on the user.

* Fix bad formulas/typos in docs, clarify SplitRange() feature of RooAbsPdf.

* Typos, misplaced comments, clarifications.

This was not causing the observed failures.

This reverts commit ede6463b.

in order to make the interactive experience in python and at the
prompt more rewarding:
```
root [0] ROOT::Math::XYZTVector v
(ROOT::Math::XYZTVector &) (0,0,0,0)
```

this commit suggests to prefer the usage of ROOT::Math::TLorentzVector
specialisations instead of the TLorentzVector class given the
advantages of the former.
- Release notes are changes
- The documentation of TLorentzVector updated to point to the specialisations of ROOT::Math::TLorentzVector
- The RDF tutorials using TLorentzVector upgraded

This reverts commit 2b071027.

Try to appease the incremental builds. They fail with:

In file included from test_Persistency1_C_ACLiC_dict dictionary payload:7:
./test_Persistency1.C:69:8: error: redefinition of 'Event'
struct Event : public DataObject  {
       ^
/.../roottest/root/tree/cache/Event.h:103:7: note: previous definition is here
class Event : public TObject {
      ^
In file included from test_Persistency1_C_ACLiC_dict dictionary payload:7:
./test_Persistency1.C:86:8: error: no member named 'm_evt' in 'Event'
    d->m_evt = d->m_run = i;

to get better performance.

This fixes ROOT-10057

/build/jenkins/night/LABEL/mac1014/SPEC/cxxmod-noimt/root/test/testGenVectorVc.cxx:194:1: error: redefinition of 'Point' as different kind of symbol
using Point = ROOT::Math::PositionVector3D<ROOT::Math::Cartesian3D<T>, ROOT::Math::DefaultCoordinateSystemTag>;
^
In module 'Darwin' imported from /Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/usr/include/assert.h:42:
/Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/usr/include/MacTypes.h:538:8: note: previous definition is here
struct Point {
^

See  http://root-forum.cern.ch/t/tselector-ttreereader-and-fchain-setentry/30170/1
and http://root-forum.cern.ch/t/problem-with-multiple-files-in-tchain-with-makeselector-class/32534/7

Without this patch for Name<Content>::Inner, MakeProject was using Name<Content,::Inner>::Inner.

in order to make of this code the most up-to-date set of examples possible.

GetEntry is not thread-safe anyway, and GetEntriesUnsafe is faster.

GetEntriesFast is not as fast as it could be: it constructs and
destructs a TReadLockGuard, and might need to modify fLast.
GetEntriesUnsafe is a thread-unsafe version of GetEntriesFast
that side-steps these operations when possible.

The operation is relatively costly as TObjArray::GetEntries also
creates a TReadLockGuard.

From V.Vasilev, it fixes:
[ 72%] Building CXX object interpreter/llvm/src/tools/clang/lib/CodeGen/CMakeFiles/clangCodeGen.dir/CodeGenModule.cpp.o
While building module 'Cling_Interpreter' imported from /.../root/core/clingutils/src/RStl.cxx:25:
In file included from <module-includes>:5:
 /.../root/interpreter/cling/include/cling/Interpreter/RuntimeUniverse.h:13:2: error: "This file must not be included by compiled programs."
 #error "This file must not be included by compiled programs."
 ^