Initialize padded sort buffers before CUB sort

[SRO-SA]

Fixes #7

Summary:

• Zero-initializes padded open/close device buffers before they are used by the parser sort path.
• Zero-initializes the temporary depth buffer before map/scan/sort operations.
• Applies the same fix to both standard JSON and JSON Lines parser paths.
• Keeps the sort range limited to the real populated open/close count.

Testing:

• I could not run CUDA or compute-sanitizer locally because my laptop does not have CUDA access.
• I reviewed the affected code paths manually.
• Please rerun the initcheck reproducer from the issue to confirm the uninitialized device-memory reads are resolved.