[clang] [clang][perf-training] Support excluding LLVM build from PGO training (PR #126876)

[Paul Kirth via cfe-commits]

ilovepi wrote:

> Just out of curiosity, is this corpus or its end product a single huge source file? If so it would also be very useful for BOLT especially instrumentation. The alternative I'm currently using is sqlite3.c

The corpus won't be a single file. We'll need C, C++, and a couple of full projects to make sure the LTO pipeline and the linker can be exercised. 

https://github.com/llvm/llvm-project/pull/126876