Fix excessive memory usage in local storage range downloads

[mneha05]

Fix excessive memory usage in local storage range downloads

Description

Fixes #2191.

The local driver's range-stream method reads the entire source file to determine its size, then yields the whole requested range as one chunk. Even a tiny range therefore allocates memory proportional to the source size, and chunk_size is ignored. The file-download method then buffers the complete range before writing.

Use os.fstat on the open descriptor for the size, seek directly to the requested offset, and read at most the requested chunk size and remaining range length. Write each chunk to the destination as it arrives. Preserve non-inclusive end offsets and existing range validation. Use the standard default chunk size when omitted or zero, reject negative chunk sizes, and stop on early EOF.

Add five regression test methods covering bounded/default chunks, exact range reads, incremental writes, negative chunk sizes, and early EOF. All five fail against the original implementation. Add a storage changelog entry.

A local tracemalloc experiment requested 1 KiB from a 64 MiB sparse file with chunk_size=128:

Implementation	Peak traced Python allocation	Output
Original	67,113,884 bytes	One 1,024-byte chunk
Fixed	6,242 bytes	Eight 128-byte chunks

These are allocations for this single experiment, not total process memory or a general benchmark.

Status

Done, ready for review. Local checks pass; upstream CI has not yet run.

Validation on Python 3.12.14 / Linux:

• pytest -q libcloud/test/storage libcloud/test/test_utils.py --ignore=libcloud/test/storage/test_list_objects_filtering_performance.py: 982 passed, 12 skipped, with six additional passing subtests. Existing deprecation warnings remain.
• All applicable prek hooks for the three changed files pass, including Black, isort, flake8, pyupgrade, and codespell.
• git diff --check passes.

Checklist

• Code linting
• Documentation — storage changelog entry
• Tests
• ICLA (required for bigger changes)

AI assistance

Generated-by: OpenAI Codex (GPT-6)

Codex assisted with investigation, implementation, regression tests, local validation, and this description. This assistance is disclosed following the ASF Generative Tooling Guidance.