feat(executor): support reading and writing csv files

[wangrunji0408]

This PR adds 2 executors: CopyFromFile and CopyToFile, which can read chunks from file and write chunks to file in CSV format (.csv, .tbl).
Given that the csv crate only supports sync file IO, we spawn the task to blocking IO threads, and transfer the chunks through async channel in tokio.
Next step we will parse the COPY statement and link it to these executors. But I found the sqlparser crate does not fully support COPY yet (sqlparser-rs/sqlparser-rs#364). Maybe we should patch this crate first (

[skyzh]

I would suggest SELECT * FROM 'xxx.csv' directly reading the csv file as table. For export, I haven't thought of that.

[skyzh]

Rest LGTM

[skyzh]

How can we know that column types when constructing the plan? Do we need to open that file before actually executing?

[wangrunji0408]

The column types can be inferred from table catalog or query results, so we don't need to open the file.
btw, I have another question: If we are going to export data to an existing file, should we append or truncate it to empty first?

[skyzh]

We should only allow export to a new file. e.g. OpenOptions::default().create_new(true).

[skyzh]

... if exporting to an existing file, we should truncate the content, as it might contain data from other tables. btw, we should warn users before really overwrite a file.