Upgrade seqio to upstream to workaround the MaxText dependency issue for inference

[gpupuck]

Default version built with MaxText image is seqio==0.0.16 which introduces error

0: Traceback (most recent call last):
0:   File "/opt/maxtext/MaxText/inference_microbenchmark.py", line 26, in <module>
0:     from jetstream.engine import token_utils
0:   File "/opt/jetstream/jetstream/engine/token_utils.py", line 24, in <module>
0:     from seqio.vocabularies import SentencePieceVocabulary
0:   File "/usr/local/lib/python3.12/dist-packages/seqio/__init__.py", line 18, in <module>
0:     from seqio.dataset_providers import *
0:   File "/usr/local/lib/python3.12/dist-packages/seqio/dataset_providers.py", line 60, in <module>
0:     @dataclasses.dataclass(frozen=True)
0:      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
0:   File "/usr/lib/python3.12/dataclasses.py", line 1258, in wrap
0:     return _process_class(cls, init, repr, eq, order, unsafe_hash,
0:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
0:   File "/usr/lib/python3.12/dataclasses.py", line 994, in _process_class
0:     cls_fields.append(_get_field(cls, name, type, kw_only))
0:                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
0:   File "/usr/lib/python3.12/dataclasses.py", line 852, in _get_field
0:     raise ValueError(f'mutable default {type(f.default)} for field '
0: ValueError: mutable default <class 'seqio.vocabularies.PassThroughVocabulary'> for field vocabulary is not allowed: use default_factory

Upgrade seqio to directly use upstream repo to workaround the issue. (This is not the perfect but workable solution)