[tool] Added a script to print occupancy info

This PR adds a script that can print out occupancy related information of a kernel.

Example usage:

./script/amd/occ.sh ./perf-kernels/06-fused-attention-fwd-transV.py

Output

LDS:  32768, num_warps:  4
VGPRS: 256 (spill: 3)
occ: 2 waves/SIMD (occ_LDS: 2, occ_vgpr: 2)
perf: 127.4 tflops

How it works

LDS usage and num_warps are extracted from the IR
VGPR usage and spills are extracted from the ISA
For now, it is assumed
- There are 4 SIMDs per CU
- Total LDS size per CU is 65536 bytes
- Total number of VGPRs per SIMD is 512

ROCm / triton