Update tokenizers submodule to 3fc1df7 (pytorch#23028)

### Summary

Update `extension/llm/tokenizers` from `1d7ca636` to
[`3fc1df71`](meta-pytorch/tokenizers@3fc1df7),
the current upstream main commit. This brings in Unicode 17 NFC
normalization fixes, reliable linking of the lookahead regex fallback,
and the trie-based special-token matcher. [Upstream
changes](meta-pytorch/tokenizers@1d7ca63...3fc1df7).

### Test plan

Debug build on Apple M1 Pro succeeded. 14/15 CTest executables passed,
including `test_normalizer` and `regex_lookahead_link_test`.
`test_hf_tokenizer` could not create its temporary fixtures in
`/var/tmp` because the local harness denies writes there; an independent
write probe confirmed `Operation not permitted`. Upstream macOS, Linux,
and Windows CI passed for the pinned commit.

```sh
cmake -S extension/llm/tokenizers/test -B /tmp/tokenizer-tests -DCMAKE_BUILD_TYPE=Debug -DCMAKE_POLICY_VERSION_MINIMUM=3.5
cmake --build /tmp/tokenizer-tests --parallel 8
ctest --test-dir /tmp/tokenizer-tests --output-on-failure
```

Authored with OpenAI Codex.

Author: mergennachin