storableToF should use target instead of host arch?

Currently we have:

```
{-# INLINE CONLIKE storableToF #-}
storableToF :: forall a. Storable a => FixedPrim a
-- Not all architectures are forgiving of unaligned accesses; whitelist ones
-- which are known not to trap (either to the kernel for emulation, or crash).
#if defined(i386_HOST_ARCH) || defined(x86_64_HOST_ARCH) \
|| ((defined(arm_HOST_ARCH) || defined(aarch64_HOST_ARCH)) \
&& defined(__ARM_FEATURE_UNALIGNED)) \
|| defined(powerpc_HOST_ARCH) || defined(powerpc64_HOST_ARCH) \
|| defined(powerpc64le_HOST_ARCH)
storableToF = FP (sizeOf (undefined :: a)) (\x op -> poke (castPtr op) x)
#else
storableToF = FP (sizeOf (undefined :: a)) $ \x op ->
if ptrToWordPtr op `mod` fromIntegral (alignment (undefined :: a)) == 0 then poke (castPtr op) x
else with x $ \tp -> copyBytes op (castPtr tp) (sizeOf (undefined :: a))
#endif
```

Isn't this broken during cross compilation? E.g. if I compile on powerpc_HOST_ARCH targeting a platform that doesn't support unaligned access wouldn't this fall over?

I'm not an expert on cross compilation but seemed wrong to me.

Locate storableToF and the architecture conditionals shown in the issue; first verify whether HOST_ARCH refers to the build or target architecture during cross compilation. Inspect or reproduce a cross-compilation setup where target alignment differs, then confirm the safe unaligned-access path is selected for the target; done means behavior is correct for that target.

Written by the indexing model from the issue text.