[GPU] Add optimized permute kernel for B↔F axis swap (order [1,0,2,3])

[andrew-k-park]

Details:

• Add optimized GPU permute kernel (permute_b_f_axes) for B↔F axis swap pattern (order [1,0,2,3] and higher-dim equivalents)
• Vectorize memory access along the contiguous X dimension using vload/vstore (16-byte transactions per work-item)
• 43 unit tests added covering various shapes, data types, and edge cases — each test verifies the new kernel produces identical results to the reference implementation

Tickets:

• CVS-182676

AI Assistance:

• AI assistance used: yes
• assist with kernel implementation, test case generation