compiler-rt: re-enable memcpy runtime safety in tests

[dweiller]

The PR re-enables runtime safety for compiler-rt memcpy tests. This should allow detecting logic bugs more reliably with tests.

The diff looks messy due to indentation changes, but the only actual changes are the re-inclusion of 3 asserts that were removed in #22854 and making the implementation generic based on a comptime enable_safety: bool parameter. This allows the tests to test the logic with enable_safety == true while not causing recursive calls to memcpy, as the memcpy calls made by safety code will call the version with enable_safety == false.

Another option would have been to have enable_safety be a comptime argument to all the helper functions - this makes a cleaner diff, but I think it makes the code a bit harder to read.

[andrewrk]

what's the problem statement?

[dweiller]

The problem I see is that having safety disabled in tests means that there is a higher chance that bugs introduced to memcpy in the future may pass the tests. Perhaps it's not likely to actually happen, but I thought it's best to be safe.