fix(storage): correct off-by-one in BlobReader chunked range fetch

[Yonghui-Lee]

Summary

Fixes an off-by-one in BlobReader.read() that caused every chunked download to request one byte more than chunk_size.

download_as_bytes(end=N) is documented as inclusive (HTTP Range semantics).
fileio.py:137 passed end = fetch_start + chunk_size, which fetches chunk_size + 1 bytes.

Why it was invisible

• Output is correct: the over-fetched byte is absorbed by the buffer and self._pos is incremented by actual bytes consumed.
• Bandwidth waste is tiny: 1 byte per chunk; on the default ~40 MiB chunk size that's 0.0000024% overhead.
• Tests passed: the test mocks used Python slice semantics (TEST_BINARY_DATA[start:end], end-exclusive), so the test mock and the production code agreed with each other.

What changed

• google/cloud/storage/fileio.py: end = fetch_start + chunk_size - 1
• tests/unit/test_fileio.py: rewrote every read_from_fake_data mock to slice [start:end+1] so it reflects real GCS HTTP Range semantics, and shifted explicit end=N assertions by -1 (e.g. end=8 → end=7).
• Added test_read_uses_inclusive_end_range to prevent silent regression: with chunk_size=64 it asserts the first fetch uses end=63, not end=64.

[gemini-code-assist]

Code Review

This pull request corrects an off-by-one error in the BlobReader.read method where the end parameter for download_as_bytes was incorrectly treated as exclusive. The fix involves subtracting 1 from the fetch_end calculation to align with HTTP Range semantics. Unit tests have been updated to reflect this change, and a new regression test was added. Review feedback suggests clarifying a comment to avoid ambiguity and using more robust mock assertions in the test suite to prevent potential TypeErrors if a mock is not called.