Enhance the metadata reading during the Copy From scenarios

[WenyXu]

What type of enhancement is this?

Performance

What does the enhancement do?

The file reader is designed for the sync reader. They typically support relatively cheap seek operation. However, if we try to seek on stream bytes (e.g., read bytes from S3), the seek position could be costly (re-send a new get request to S3).

I checked CSV and JSON format, and these file format reader only requires Read trait to infer a schema. Therefore, we might only need to care about the Parquet and the ORC format.

See also: #3191

Implementation challenges

No response