What this PR does / why we need it:

pa_to_feast_value_type only parsed the list<item: T> spelling of a PyArrow list type. Parquet files written by pandas/pyarrow read back as list<element: T>. So for a FileSource column typed Array(String), Array(Bytes) or Array(UnixTimestamp), schema inference in feast apply fails, and so do write_to_offline_store and push(..., to=PushMode.OFFLINE) (both go through get_pyarrow_schema_from_batch_source). They raise:

KeyError: 'list<element: string>'

Repro:

df = pd.DataFrame({"user_id": [1, 2], "tags": [["a", "b"], ["c"]], "event_timestamp": [ts, ts]})
df.to_parquet("tags.parquet")  # tags: list<element: string>
src = FileSource(path="tags.parquet", timestamp_field="event_timestamp")
store.apply([user, src, FeatureView(name="tags", entities=[user], source=src)])  # KeyError

Integer, float and bool lists only worked because #5118 hard-coded some list<element: ...> keys. Its string and bytes keys (list<element: str>, list<element: bytes>) use names pyarrow never prints, and there was no timestamp key. large_list<...>, list<item: T not null> and element-named nested lists raise the same KeyError.

Changes:

• One pattern, (large_)list<(item|element): T( not null)>, now parses the list prefix. The recursion into T and the rule that nested lists map to VALUE_LIST are unchanged. The hard-coded list<element: ...> keys are removed.
• _parse_pa_type_str in offline_utils.py, which rebuilds nested list types for write_to_offline_store, uses the same helper. list<element: list<element: int64>> now becomes list<list<int64>> instead of falling back to string.

Every type string that mapped before still maps to the same ValueType, except the two keys pyarrow never produces.

Which issue(s) this PR fixes:

No open issue; repro above.

Checks

• I've made sure the tests are passing.
• My commits are signed off (git commit -s)
• My PR title follows conventional commits format

Testing Strategy

• Unit tests
• Integration tests
• Manual tests
• Testing is not required for this change

Misc

Unit: TestPyArrowListTypeStrings in tests/unit/test_type_map.py covers parametrized spellings for pa_to_feast_value_type and _parse_pa_type_str, plus a tmp_path test that writes Parquet with df.to_parquet, reads the schema through FileSource, and checks both the inferred ValueType and get_pyarrow_schema_from_batch_source. 13 of the 18 new cases fail on master; the other 5 pin spellings that already worked.

Manual: a local repo with a FileSource over a df.to_parquet file, with string, bytes, timestamp, int, float, bool and nested list columns. On master, feast apply with inference, write_to_offline_store (file and DuckDB offline stores) and push(to=PushMode.OFFLINE) (DuckDB) all raise the KeyError. With this change they succeed, and materialize plus get_online_features return the lists.