[python] Align SnapshotManager constructor with Java

[TheR1sing3un]

Purpose

Switch the Python SnapshotManager.__init__ from (table) to the Java-aligned (file_io, table_path, branch=None, snapshot_loader=None), so the class no longer depends on FileStoreTable. Mirrors paimon-core/.../utils/SnapshotManager.java's constructor (which takes FileIO, Path tablePath, @Nullable String branchName, @Nullable SnapshotLoader, @Nullable Cache); the Cache parameter is left out of scope (Caffeine has no Python equivalent yet).

The previous Python signature held self.table = table and reached through self.table.file_io / self.table.catalog_environment.snapshot_loader() / self.table.current_branch() to populate fields. That made SnapshotManager a layering-violation -- the snapshot layer reaching back into the table layer -- and forced mock tests to stub a four-attribute mock table just to construct the manager. The new constructor takes the four basics directly.

In scope

• pypaimon/snapshot/snapshot_manager.py: new constructor; self.table field removed; copy_with_branch rewritten to construct the rebranched manager via the new constructor (no field-swap).
• pypaimon/table/file_store_table.py: snapshot_manager() remains the canonical factory and now wires (file_io, table_path, current_branch(), catalog_environment.snapshot_loader()) directly.
• 11 production call sites + 7 raw-construction test sites migrated to table.snapshot_manager().
• Mock-style tests in streaming_table_scan_test.py / file_store_commit_test.py / partition_predicate_test.py: drop the @patch('...SnapshotManager') decorator and the mock parameter; wire each test's mock_snapshot_manager through table.snapshot_manager.return_value, matching how production code obtains its instance.
• snapshot_manager_test.py: simplified mock setUp -- no more layered `Mock()` table, just `SnapshotManager(file_io, "/tmp/test_table")`.

Out of scope

• Caffeine-style `Cache<Path, Snapshot>` parameter (Java-only; Python Cache port is a separate effort).
• Other Java SnapshotManager API gaps (latest_snapshot_id / earliest_snapshot_id / snapshot_exists / etc.) -- separate follow-up PRs.

Tests

From `paimon-python/`:

```
pytest pypaimon/tests/snapshot_manager_test.py pypaimon/tests/branch_manager_test.py \
pypaimon/tests/streaming_table_scan_test.py pypaimon/tests/file_store_commit_test.py \
pypaimon/tests/partition_predicate_test.py -q # 76 passed
pytest pypaimon/tests/ --ignore=pypaimon/tests/rest --ignore=pypaimon/tests/e2e_rest \
--ignore=pypaimon/tests/sql_context_test.py # 963 passed
flake8 --config=dev/cfg.ini # clean
```

(`sql_context_test.py` requires the optional `datafusion` extra which isn't installed in this environment; failures there are pre-existing on master and unrelated to this diff.)

Anti-divergence checklist

• Constructor parameter order matches Java `SnapshotManager(FileIO, Path, @nullable String, @nullable SnapshotLoader, ...)`.
• `branch == "main"` keeps existing paths bit-identical (`BranchManager.branch_path(p, "main") == p`); call sites that passed `current_branch() == "main"` get zero-regression behaviour.
• `copy_with_branch` mirrors Java's factory: new `(file_io, table_path, branch_name, snapshot_loader.copy_with_branch(...) if loader else None)`.

Generative AI disclosure

Drafted with assistance from a generative AI tool. All code, tests, and Java alignment were reviewed and validated by the contributor.

---

Note: built on top of #7756. Once #7756 merges, this branch will rebase cleanly onto master.