Why did the compiler even chose to fetch DWORDs only in the first place? It's unclear to me why the accumulator (apparently) determines the vectorization width?
Sure there's a gazillion ways of detecting a debugger (specifically on windows), but then we are back to detecting debuggers. My point was that VEH (alone) doesn't prevent any sort of debugging, specifically not the VEH handler itself.
Btw, debuggers (on windows) won't usually install VEH to support BPs, they'll use the win32 debugger infrastructure where the OS manages exceptions and delivers them to the attached debugger object (which again can be detected in several ways).
They also do not technically need HW bp registers, although often they will. A simple way to implement BPs is to write 0xCC (INT3) to the text section, then restore original bytes when the INT3 fires.
A debugger sees exceptions before the application does, and it knows which exceptions it should handle by itself (e.g. breakpoints set by the user) vs passing them to the application.
I don't have a windows machine rn to verify but I'd expect your assumption that it's impossible to debug an INT3 VEH to be incorrect.