Rewrite the RDataFrame JIT logic to avoid controlled leaks

[vepadulano]

Explain what you would like to see improved and how.

The RDataFrame JIT logic makes use of a series of controlled memory leaks between different functions to ensure lifetime of objects that are needed by the JIT until that actually happens.

This is an optimization but has certain fragilities, that were already analysed in a recent bug report. The same topic surfaced in the past (see this issue moved from JIRA). And the fact we rely on leaking objects is subtly error-prone (see a recent PR).

We should investigate how to implement a different way to handle the JIT machinery so that it does not require leaking objects.

ROOT version

Any

Installation method

Any

Operating system

Any

Additional context

No response