[compile_time] Break-down of compilation time for Triton

[xuzhao9]

We are interested in adding Triton compilation time breakdown on Tritonbench, roughly there are three parts:

• Aggregated time spent on Triton autotuning;
• Time spent on PTXAS compilation;
• Time spent on compiling Triton code to PTXAS.

Ideally we can get these metrics from Proton. If not, we can only measure the E2E compilation time.

[xuzhao9]

Re-opening since we can now get by-stage compile time from Kineto trace