Make O2 the default optimization level for aie-opt

[newling]

Relevant performance numbers scraped from CI. These are the direct codegen benchmarks that change from O3 to O2.

Time Comparison

Benchmark	Before (us)	After (us)
matmul_512_4096_512_bf16_f32_npu1_4col_outline	2469.0	2483.0
matmul_transpose_b_512_4096_512_bf16_f32_npu1_4col_outline	1948.0	1950.0
matmul_4096_512_512_bf16_f32_npu1_4col_outline	1893.0	1916.0
matmul_transpose_a_4096_512_512_bf16_f32_npu1_4col_outline	2422.0	2332.0
matmul_512_4096_512_bf16_f32_npu1_4col_outline_4_level_tiling	1388.0	1615.0

Memory Comparison

Benchmark	Before (bytes)	After (bytes)
matmul_512_4096_512_bf16_f32_npu1_4col_outline	8464	3504
matmul_transpose_b_512_4096_512_bf16_f32_npu1_4col_outline	9216	3536
matmul_4096_512_512_bf16_f32_npu1_4col_outline	8464	3504
matmul_transpose_a_4096_512_512_bf16_f32_npu1_4col_outline	8656	3520
matmul_512_4096_512_bf16_f32_npu1_4col_outline_4_level_tiling	14816	3184

So matmul_512_4096_512_bf16_f32_npu1_4col_outline_4_level_tiling needs some investigation.

[yzhang93]

Maybe leave this test matmul_512_4096_512_bf16_f32_npu1_4col_outline_4_level_tiling using O3 for now?

[newling]

Maybe leave this test matmul_512_4096_512_bf16_f32_npu1_4col_outline_4_level_tiling using O3 for now?

I think the O3 code is better than O2 for all the tests, it's just that this is only test where the data movement is fast enough for it show up. So I'm going to dig deeper to see what else I can do to make O2 match O3. Ideally the code generated with the 2 should be identical.