[ntuple] Grow RVec capacity geometrically in ResizeRVec

[ishaaaan17]

this pull request:

makes RRVecField::ResizeRVec grow capacity geometrically (doubling, capped at std::int32_t max) when reading variable-sized ROOT::RVec<T> collections.

before, ResizeRVec reallocated just enough for nItems. if the collection size crept up from event to event, that meant a lot of malloc/free cycles.

this takes care of the todo in tree/ntuple/src/RFieldSequenceContainer.cxx:
// TODO Increment capacity by a factor rather than just enough to fit the elements.

changes or fixes:

• when an RVec that is already allocated needs to grow, its capacity now grows geometrically instead of to exactly nItems.
• the first allocation and memory-adopting transfers still use the exact capacity needed, so single-element containers don't get over-allocated.
• added the unit test TEST(RNTuple, RVecGeometricGrowth) in tree/ntuple/test/rfield_vector.cxx. it checks the growth and that capacity is kept across entries of different sizes.
• formatted with ROOT's .clang-format.

checklist:

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

ai-assisted coding disclosure:

Assisted-by: antigravity:gemini. i used an llm assistant to look at different ways of growing the capacity and to draft the first version of the changes. i then went through the memory semantics, edge cases, alignment constraints and the tests myself.

[silverweed]

Seems reasonable to me. Although should we also provide a ShrinkToFit method for specific cases that want to reclaim all possible memory?
@jblomer thoughts?

[github-actions]

Test Results

    24 files      24 suites   3d 23h 5m 46s ⏱️
 3 886 tests  3 885 ✅ 0 💤 1 ❌
83 181 runs  83 180 ✅ 0 💤 1 ❌

For more details on these failures, see this check.

Results for commit 3ba182d.

[jblomer]

Thank you very much for addressing this TODO!

I took your commit and extended it a it (since we are on the subject) in PR #23718

Specifically, I changed the growth factor to 1.5 for all but small vectors (with the aim to be a bit more memory-friendly). An in the same spirit, there is also automatic shrinking added.

Let's continue the discussion in #23718

[jblomer]

Thank