octets: fix slice_last() with non-zero offset

[TerryTaoYY]

Summary

Fix Octets::slice_last() and OctetsMut::slice_last() returning an incorrect tail slice when the internal offset has been advanced.

These helpers previously used cap() (remaining bytes after off) to compute the start index, but then applied that index to the full underlying buffer (self.buf[...]). When off > 0, this could include already-consumed bytes and return the wrong region.

Reproduction

let mut b = Octets::with_slice(b"helloworld");
b.get_bytes(5).unwrap(); // advance offset past "hello"
assert_eq!(b.slice_last(4).unwrap(), b"orld");

Before this change, slice_last(4) could return the wrong region (e.g. starting at index 1), because cap() was treated as an absolute index into the full buffer.

Fix

Compute end as self.buf.len() and slice self.buf[end - len..end] (and the mutable equivalent). The existing len <= self.cap() guard ensures end - len >= off, so the returned slice stays within the remaining window.

Tests

Add regression cases that advance the offset via get_bytes() and then assert slice_last() returns the correct tail bytes for both Octets and OctetsMut.

Tested with:

• \ cargo test -p octets

[ghedo]

Merged, thanks @TerryTaoYY!