This works as expected for reading:
fn u64_from_low_eight(buf: &[u8; 9]) -> u64 {
    let bytes: &[u8; size_of::<u64>()] = buf[..size_of::<u64>()].try_into().unwrap();
    u64::from_le_bytes(*bytes)
}
(It nicely optimises into a single assembly instruction on AArch64 and x86_64.)
I had expected something similar to work for an unaligned write of a u64 to a buffer.
/// Encodes a u64 into 1-9 bytes and returns the number of bytes updated.
pub fn encode(value: u64, buf: &mut [u8; 9]) -> usize {
    let low64: &mut [u8; size_of::<u64>()] = &mut buf[..(size_of::<u64>())].try_into().unwrap();
    match value {
        // FIXME: Change to exclusive ranges once the feature's stabilised.
        OFFSET0..=OFFSET1_LESS_ONE => {
            let num = inner_encode::<1>(value, low64);
            #[cfg(test)] eprintln!("low64: {low64:?}");
            #[cfg(test)] eprintln!("buf: {buf:?}");
            num
        },

use std::mem::size_of;
fn u64_to_low_eight(value: u64, buf: &mut [u8; 9]) {
    let low64: &mut [u8; size_of::<u64>()] = &mut buf[..size_of::<u64>()].try_into().unwrap();
    *low64 = u64::to_le_bytes(value);
    dbg!(low64);
}
fn main() {
    let mut src: [u8; 9] = [1, 2, 3, 4, 5, 6, 7, 8, 9];
    u64_to_low_eight(0x0A_0B_0C_0D_0E_0F_10_11, &mut src);
    dbg!(src);
}

   Compiling playground v0.0.1 (/playground)
    Finished dev [unoptimized + debuginfo] target(s) in 0.62s
     Running `target/debug/playground`
[src/main.rs:6] low64 = [
    17,
    16,
    15,
    14,
    13,
    12,
    11,
    10,
]
[src/main.rs:12] src = [
    1,
    2,
    3,
    4,
    5,
    6,
    7,
    8,
    9,
]

// 获取对这8个复制字节的可变引用
let low64: &amp;mut [u8; size_of::<u64>()] = &amp;mut (
    // 从`buf`中复制8个字节
    buf[..size_of::<u64>()].try_into().unwrap()
);

let low64: &amp;mut [u8; size_of::<u64>()] = (&amp;mut buf[..size_of::<u64()>]).try_into().unwrap();