[Clang] Incorrect overflow check and evaluation for MS integer literal suffixes (i8/i16/i32/i64)

Inspired by #212504

I discovered other issues while trying to fix this assertion.

For code : `static_assert(-255i8, "");` Surprisingly, the current mainline Clang doesn't produce any errors or warnings at all.

The current Clang mainline produces neither errors nor warnings for this, yet the value is actually truncated; consequently, it fails the assertion `static_assert(-255i8 == -255, "-255 != -255 ?")`.

https://godbolt.org/z/K88b1sxz9

This drew my attention to the fact that the logic handling Microsoft-style integer literal extensions appears quite fragile. I plan to submit a patch later, though some edge cases may still require further discussion.

For example, how should `-128i8` be handled?

Reproduce the two static_assert examples and inspect Clang's handling of Microsoft-style i8/i16/i32/i64 integer literal suffixes. Clarify the intended behavior for boundary cases such as -128i8, then ensure overflow, truncation, diagnostics, and constant evaluation are consistent.

Written by the indexing model from the issue text.