[python] Implement partition expiration for pypaimon

[JunRuiLee]

Purpose

Add pypaimon support for expiring partitions based on table options and CLI overrides, matching the Java partition expiration behavior for values-time and update-time strategies.

This includes:

• Partition expiration orchestration for FileStoreTable.
• Partition entry discovery from manifest metadata.
• Time extraction from partition values.
• Java DateTimeFormatter-style timestamp formatter support.
• table expire-partitions CLI command.
• Partition expiration options in Python CoreOptions.
• Null partition handling via partition.default-name.
• Default partition.expiration-max-num=100.

Tests

Added paimon-python/pypaimon/tests/partition_expire_test.py covering:

• Extracting partition time from simple date and timestamp values.
• Extracting time with timestamp patterns across multiple partition fields.
• Java formatter conversion cases like yyyyMMdd.
• Selecting expired partitions with values-time.
• Selecting expired partitions with update-time.
• Skipping unparseable partition values.
• Respecting expiration check interval.
• Respecting max expire count.
• Empty partition lists.
• Unconfigured expiration.
• Non-partitioned tables.
• Duration parsing for days, hours, minutes, seconds, milliseconds, and combined durations.

[JingsongLi]

Review: [python] Implement partition expiration for pypaimon

Good work porting the partition expiration logic to pypaimon. The architecture is clean and the strategy pattern mirrors the Java implementation well. A few issues I noticed:

---

Bug: Dead code in _parse_with_formatter (partition_time_extractor.py)

The except block in _parse_with_formatter calls datetime.strptime(timestamp_string, formatter) again — the exact same call that just raised ValueError. This will always re-raise. The Java version uses a different formatter in the fallback (LocalDate.parse vs LocalDateTime.parse). To match the Java behavior, the fallback should attempt date-only parsing with a separate format.

---

Design: Java formatter conversion (_JAVA_TO_PYTHON_PATTERNS) is fragile

1. SSS/SS -> %f: Python's %f expects 6 digits (microseconds), while Java's SSS is 3-digit millis. This is a subtle semantic mismatch.
2. Ordering risk: Sequential str.replace() can produce incorrect results for overlapping tokens.

Consider a regex-based tokenizer or documenting the constraints.

---

Correctness: PartitionUpdateTimeExpireStrategy with last_file_creation_time=0

When last_file_creation_time is 0, the partition will always be considered expired. Consider adding a guard or log warning.

---

Scope: Unrelated methods in file_store_table.py

The statistics() and analyze() methods appear unrelated to partition expiration and should be in their own PR.

---

Minor issues

1. _parse_duration type hint says str but tests pass None — add Optional[str].
2. No expire-batch-size support (Java has this to prevent OOM on large drops).
3. Verify truncate_partitions has same semantics as Java's commit.dropPartitions(...).
4. Missing .done partition handling for metastore-managed tables.

---

Tests

Thorough coverage. Suggest adding a test for last_file_creation_time=0 to document the "always expired" behavior.

Overall solid work. The main actionable fix is the _parse_with_formatter bug.

[JunRuiLee]

Thanks @JingsongLi for the review, I've addressed all comments. PTAL~