in order to take into account differences arising when one of the two
typeinfos come from a jitted context.

The callback mechanism as implemented is not what is needed by CMS,
as previously thought. In order to allow the final goal of avoiding
oversubscription of the machines on which the TBufferMerger is used
within a framework, this functionality is better removed in order to
allow doing away also with the separate output thread currently needed
by TBufferMerger.

In line with what it's done for jitted filters, now the jitting
of custom columns creates the RCustomColumn and assigns it as the
fConcreteCustomColumn of the corresponding RJittedCustomColumn that
was previously booked.
This avoids having the situation in which a certain custom column
has been "booked" but is not yet present in the map of custom columns.

It will also help with the "local custom columns" that are coming
with ROOT-9465.

We jit actions as well as Define's and Filter's now.

They were added when custom columns where nodes of the computation
graph, but are useless now that they are not anymore.

See e.g. C++ core guidelines C.128

this prevents bogus behaviour if two instances of TSlotStack are used simultaneously
or even one after the other.

`TPaveText` containing the title was not correct.

the full method was protected with the global lock. Now we split it
in two parts, the first one where the cache is looked at and no state
is modified with a read lock. The second, where the cache is filled
and the interpreter is invoked with a write lock.

As suggested by Stefan Wunsch, decorators provide a nicer syntax
to mark that a function implements some pythonization.