schema evolution for ROOT I/O

[chenel]

As there are more and more uses for CAFs, changes to the format become increasingly disruptive, since CAFs made with differing versions of duneanaobj have to be read back with a version that's compatible with their structure.

ROOT's "schema evolution" concept can be used to deal with many of these cases. We should begin using it to help reduce the number of "breaking changes" that adjustments to the storage format cause.

[chenel]

Notes from Mar 20 2025 CRAB meeting:

• good contacts for experience / advice to follow up with: Peter Van Gemeren, Philippe Canal, Erica Snider, Kyle Knoepfel
• need to make sure schema evolution works invisibly, or failing that, have appropriate documentation, so users can do side studies to show their favorite variable addition improves some analysis output
• How does this intersect with RNTuple?
• should try to work out in test case (e.g., MaCh3) and report proposed solution to CRAB for further feedback

[chenel]

Tom Junk also reminds me that we will also need to think about:

• how srproxy will handle interact with this system
• what will happen when trying to read back "flat" CAFs