Disk-backed large dataset support via SQLite temp storage

[vmvarela]

Part of #69
Depends on #69-spike (evaluate streaming approaches)

Description

Implement the simpler streaming approach: configure SQLite to use disk-backed temp storage for large datasets. This allows all SQL operations to continue working on datasets larger than RAM, without changing query semantics.

Acceptance Criteria

• --memory-limit <size> flag sets a hint for SQLite temp storage threshold (e.g. 256MB, 1GB)
• SQLite configured with PRAGMA temp_store = FILE when memory limit is set
• Temp files are cleaned up after execution
• Queries that previously ran out of memory on 1GB+ files now complete successfully
• Existing behavior unchanged when flag is not set
• Error message if temp directory is not writable

Notes

• Use PRAGMA temp_store_directory to configure temp location
• No changes to query parsing or output — only storage backend changes
• Follow spike recommendation on exact PRAGMA configuration