Fix from raw parts

[yilin0518]

Objective

AlignedVec::from_raw_parts's safety documentation has a contract gap: when capacity == 0, it only said ptr "need only be aligned" and never required align itself to be a non-zero power of two. This means a caller can construct an AlignedVec that satisfies the documented contract literally (e.g. align = 3, capacity = 0), yet later trigger undefined behavior through purely safe method calls (e.g. push), because internal methods like layout() unconditionally build a Layout via Layout::from_size_align_unchecked(self.cap, self.align), which requires align to be a non-zero power of two regardless of capacity.

The old docs were also imprecise about:

• there being no upper bound stated on align independent of allocation
• what "capacity ... needs to be the same size" meant when the pointer isn't required to be allocated
• what exact byte alignment ptr must satisfy when capacity == 0

Solution

Rewrote the # Safety section of from_raw_parts to state the invariants unconditionally and precisely:

• align must be a non-zero power of two, and less than isize::MAX — regardless of capacity, since align is used to build a Layout even when the vector currently has zero capacity.
• capacity, rounded up to the nearest multiple of align, must not overflow isize.
• If capacity != 0, ptr must have been allocated with a Layout using this exact align and a size of capacity (merging the two previously-separate, overlapping clauses about matching align/size on dealloc).
• If capacity == 0, ptr need not point to allocated memory, but must still satisfy ptr.as_ptr() as usize % align == 0 (replacing the ambiguous "need only be aligned").

No functional code changed — this is a documentation-only fix to close the public unsafe contract gap.

Testing

• Verified the change compiles: cargo doc -p bevy_platform --no-deps passes, and the existing doctest on from_raw_parts still passes.
• To confirm the gap in the original docs was real (not just a theoretical wording nitpick), I built a standalone reproduction outside this repo:

use bevy_platform::collections::AlignedVec;
use core::ptr::NonNull;

fn test_align_not_power_of_two() {
    unsafe {
        let ptr = NonNull::new(3usize as *mut u8).unwrap();
        let mut vec = AlignedVec::from_raw_parts(ptr, 3, 0, 0);
        // Triggers UB: Layout::from_size_align_unchecked(new_cap, 3)
        // requires align to be power of 2.
        vec.push(1);
    }
}

fn main() {
    test_align_not_power_of_two();
}

I use the command cargo +nightly-x86_64-pc-windows-gnu miri run to test the above code, and it trigger the unsafe precondition(s) violation:

thread 'main' (1) panicked at D:\projects\bevy\crates\bevy_platform\src\collections\aligned_vec.rs:227:30:
unsafe precondition(s) violated: Layout::from_size_align_unchecked requires that align is a power of 2 and the rounded-up allocation size does not exceed isize::MAX

This indicates a bug in the program. This Undefined Behavior check is optional, and cannot be relied on for safety.
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
note: in Miri, you may have to set `MIRIFLAGS=-Zmiri-env-forward=RUST_BACKTRACE` for the environment variable to have an effect
thread caused non-unwinding panic. aborting.