[core] support decouple the delta files lifecycle

[Aitozi]

Purpose

This PR is meant to support decouple the delta files lifecycle #2899

The basic idea behind this is that:

• Add fileSource in DatafileMeta to indicate whether this file is generated as an APPEND or COMPACT file
• When expire a snapshot, if changelog is decoupled, and it's a none changelog producer, we do not clean the APPEND files in data file
• When expire a changelog, we do the final clean for these files
• Due to the manifest file could still be used by delta files, so the base and delta manifest file for the none producer are also postpone to delete

About why we need FileSource in DataFileMeta

For none changelog producer, only APPEND commits are required for stream read. In a COMPACT commit, some files from the compact or append could be marked as delete. We should delete the files from the compact commit and keep the files from the append commit for further stream read. So we need a flag to distinguish the file source (compact or append).

Linked issue: close #xxx

Tests

API and Format

Introduce FileSource in DataFileMeta

Documentation

[JingsongLi]

Perhaps it's best for us to have an abstract mechanism to ensure that we don't keep both the changelog and delta files at the same time, so that we can better understand this set of things.

[Aitozi]

Perhaps it's best for us to have an abstract mechanism to ensure that we don't keep both the changelog and delta files at the same time, so that we can better understand this set of things.

I extract this logic to SnapshotDeletion and ChangelogDeletion to handle the manifestList and data file clean.

In this PR, I also refactor the ExpireChangelogImpl and ExpireSnapshotImpl to immutable, So that, we could know whether the changelog decouple is enabled in one entry point in ExpireSnapshots.Expire

[Aitozi]

Before, if user not specify the retain_min, the default value is 1 in ExpireSnapshotsImpl, now the default value is fallback to the CoreOptions.SNAPSHOT_RETAIN_MIN = 10, so if max is 2, we should manually specify the retain_min => 1. I think the current behavior is more consistent, I'm not sure whether this will break the compatibility. Please also help check this cc @JingsongLi

[JingsongLi]

Already merged to three PRs.