perf: optimize instruction_lookups::generate_witness() with improved parallelism

[crStiv]

Fix #293

• Add parallel processing for subtable_lookup_indices
• Pre-allocate memory for vectors to reduce reallocations
• Optimize memory usage pattern in processing polynomials
• Improve data unpacking efficiency with better vector capacity management