[clang] [clang] Accept mtime change for input files of relocatable PCH (PR #196057)

[Arseniy Zaostrovnykh via cfe-commits]

necto wrote:

 Hi @jansvoboda11, sorry for this unprompted PR. You seem to be the most active in this file. I am not sure who I should ask for a review, so feel free to redirect to someone else.

Essentially with this fix I shift the PCH invalidation heuristics from FPs to FNs: if previously it could often reject a PCH because of relocated inputs, now it would more often accept it, and might accept even if the file content changes (if `ValidateASTInputFilesContent` is not set). I feel this is less surprising, but I am not experienced in this part of the code, so feel free to challenge this.

This patch came about in my work with Cross-TU symbolic execution using Clang Static Analyzer.

I recommend reviewing by commit: first one documenting the status quo with a test case, second one implementing the fix and demonstrating its effect in the test case.

https://github.com/llvm/llvm-project/pull/196057