BytesCursor abstraction in FuseDevWriter

[stepancheg]

FuseDevWriter.buf field is ManuallyDrop<Vec<u8>>, but actually it is a pair of &mut [u8] space and position.

Express this logic more explicitly with

pub(crate) struct BytesCursor<'a> {
    slice: &'a mut [u8],
    position: usize,
}

Because:

• less unsafe code
• more operations are checked
• with this change it is clear that there's no UB (for example, related to aliasing &mut pointers). It may be not an issue, but with this change is much more clear

[bergwolf]

somehow CI is not running. Let me close and reopen to trigger it.

[jiangliu]

Nit: clippy (byte_char_slices) flags this and CI runs with -Dwarnings:

    b.extend_from_slice(b"ab");

[jiangliu]

Nit: the sibling call site in write_from uses self.buf.written()[..cnt]. Both are equivalent here (position was 0 before the write), but written()[..cnt] expresses the intent more clearly — could use it here too for consistency?

  Self::do_write(self.fd, &self.buf.written()[..cnt])

[jiangliu]

Just a heads-up: FuseDevWriter is a public type, so removing PartialEq/Eq is technically an API-breaking change. I checked nothing in this repo relies on it, but it's probably worth a one-liner in the changelog.
Or we may keep it as is.

[jiangliu]

Please help to add copyright header?