Find default_path lazily in env_nvidia_tool

Set default build type to Release, and add back /Zi /RTC1 to TritonRe…

…lBuildWithAsserts

Enable FP8 conversion on sm < 89

Reduce copied headers

Add _CRT_SECURE_NO_WARNINGS

Add _CRT_SECURE_NO_WARNINGS

Bundle full CUDART and CUPTI when building Proton

Update BUILD.md

Fix when multiple processes create __triton_launcher.pyd in parallel

Set close_fds=True when calling ptxas