Fix sparse realization with vector arguments

[LeonidasZhak]

Summary

Fix realization of structurally sparse delayed operations with vector-like arguments.

Thanks to maintainers

Thanks to the DelayedArray and SparseArray maintainers for maintaining these sparse realization interfaces.

Issue or motivation

Closes #123. Adding a zero vector to a zero-valued ConstantArray is structurally sparse, but realization currently fails.

Root cause

is_sparse() identifies the zero-preserving operation, while extract_sparse_array() passes the sparse block and vector directly to SparseArray. SparseArray does not support this + operation between a sparse object and a vector.

Change

When vector-like arguments are present, evaluate the block through extract_array() and convert the result back to SVT_SparseArray. The no-argument sparse fast path is unchanged. A regression test covers the ConstantArray case.

Tests

• Full DelayedArray RUnit suite: 43 test functions, 0 errors, 0 failures.
• HDF5Array reproduction using writeHDF5Array(): passed.
• R CMD build .: passed, including all three vignettes.
• R CMD check --no-manual --no-vignettes: passed with 1 WARNING and 2 NOTEs.

Scope

This does not change dense extraction, the existing no-argument sparse path, or the SparseArray API.