encode_with_unstable does not handle surrogate pairs like encode

Encoding.encode and Encoding.encode_ordinary already handle surrogate pairs/lone surrogates by retrying after UTF-16 surrogate repair.

However, Encoding.encode_with_unstable currently raises UnicodeEncodeError for the same inputs.

Repro:

import tiktoken

enc = tiktoken.get_encoding("cl100k_base")

enc.encode("\ud83d\udc4d")  # works, same as "👍"
enc.encode_with_unstable("\ud83d\udc4d")  # raises UnicodeEncodeError

enc.encode("\ud83d")  # works, same as "�"
enc.encode_with_unstable("\ud83d")  # raises UnicodeEncodeError

Start by comparing the encode_with_unstable implementation with Encoding.encode and Encoding.encode_ordinary, focusing on their handling of UTF-16 surrogate pairs and lone surrogates. Add regression coverage for the paired and lone-surrogate examples in the issue, then run the relevant test suite and confirm encode_with_unstable matches the existing methods without raising UnicodeEncodeError.