Table properties override catalog configuration when constructing FileIO

When a table is loaded, its metadata properties are merged into the property map used to construct the table's `FileIO`, and they take precedence over the operator's catalog configuration.

`Catalog._load_file_io` computes:

```python
load_file_io({**self.properties, **properties}, location)
```

where `properties` is `metadata.properties`. Because table properties come last, a value stored in a table overrides the same key configured on the catalog. The REST path does the same at `_response_to_table` / `_response_to_staged_table`, and Glue, Hive, SQL, DynamoDB, BigQuery and `StaticTable.from_metadata` all funnel table metadata into `FileIO` construction the same way — 12 call sites in total.

The keys this reaches include implementation selection (`py-io-impl`, `s3.retry-strategy-impl`) and transport configuration (`s3.endpoint`, `s3.proxy-uri`, `s3.signer` / `s3.signer.uri`, `gcs.service.host`, `hf.endpoint`, the ADLS storage authorities). These are deployment concerns — an operator sets them on the catalog — but any principal who can commit to a table can currently override them for everyone who reads it.

Issue investigation generated via claude, reviewed by Sung, Kevin, Fokko.

從 Catalog._load_file_io 開始，追蹤 REST 的 _response_to_table 和 _response_to_staged_table 路徑，然後檢查 Glue、Hive、SQL、DynamoDB、BigQuery 和 StaticTable.from_metadata 對應的呼叫點。驗證中繼資料屬性如何合併到 FileIO 建構中，並新增覆蓋測試，證明在所有受影響的路徑上，catalog 設定對 deployment 設定仍具權威性。

由索引模型根據 Issue 內容生成。