[df] RNTuple snapshot + TTree-specific options

[hahnjo]

Follow-up from #19599 (comment): RNTuple snapshotting currently does not support fAutoFlush, which doesn't really map to RNTuple's target cluster size. We need to discuss how and if we want to support it.

Note that UntypedSnapshotTTreeHelperMT is also questionable in that regard because it writes / flushes output trees in FinalizeTask. So for RDataFrame snapshotting to TTree in MT mode, it's already more of an upper guarantee, but I guess many cluster will actually be smaller than the user wants them...

[enirolf]

I propose to expand the discussion to RSnapshotOptions and how we should handle it as a whole. They can be split up into Snapshot-specific options (fLazy, fMode, fOutputFormat, fOverwriteIfExists, fVector2RVec) and TTree-specific options (fAutoFlush, fBasketSize, fSplitLevel). fCompression(Algorithm|Level) is a bit of a special case, because it is technically TTree-specific but they can (and are) also be trivially applied to RNTuple.

The Snapshot-specific options are as far as I'm aware all accounted for for RNTuple, but obviously the TTree-specific ones are not. In addition, there is currently no way to set RNTuple-specific writing options. The latter could be resolved by adding fNTupleWriteOptions which is a RNTupleWriteOptions.

fAutoFlush directly maps to TTree::SetAutoFlush, so I'm unsure if this is something we even want approximate for RNTuple, since it is not a write option to begin with. My proposal would rather be to be more explicit in which options are TTree-specific and which are RNTuple-specific, and warning users if they set one of either while fOutputFormat is set to the other.

[hahnjo]

My proposal would rather be to be more explicit in which options are TTree-specific and which are RNTuple-specific, and warning users if they set one of either while fOutputFormat is set to the other.

This makes perfect sense to me. I've changed the issue title accordingly to include TTree-specific options in general.