[ntuple] Preparations for introducing slow merging in the RNTupleMerger

[silverweed]

This Pull request:

refactors the RNTupleMerger to modularize a bit better the parts that depend on the merging strategy (slow or non-slow, aka L4 or L1/2/3).

A new interface, RNTupleMergeStrategy, is introduced which provide the common functions that both strategy need to implement (currently only the non-slow strategy exists, called RNTupleMergeStrategyDefault).
Most of the code currently belonging to the RNTupleMerger class is moved into this strategy class, with no functional changes. The only actual changes in these functions are:

• the RNTupleMergeData struct is removed as it's being replaced by the Strategy class itself (all its fields now belong to the Strategy);
• the fNumDstEntries field of the mergeData is removed altogether because I realized it's not needed (replaced by destination.GetNEntries().

A later commit will introduce the slow merging strategy, with minimal impact on the existing code thanks to this change.

Checklist:

• tested changes locally
• updated the docs (if necessary)

[github-actions]

Test Results

    23 files      23 suites   3d 19h 17m 34s ⏱️
 3 880 tests  3 879 ✅ 0 💤 1 ❌
80 409 runs  80 408 ✅ 0 💤 1 ❌

For more details on these failures, see this check.

Results for commit 3612562.