Support full paths instead of only file-names in sidecar actions's path field

[sebastiantia]

Please describe why this is necessary.

Currently, the path field in sidecar actions only supports file names (without the scheme or parent directories). This limitation prevents the use of full URIs in the path, which implementations of delta may choose to use based on the PROTOCOL.

This change is necessary to support tables that have the v2Checkpoint reader/writer feature enabled with sidecar action path's that use file paths instead of the file name.

Describe the functionality you are proposing.

Parse the sidecar action path field (which may be a full file-path) for the file name as all sidecars should live in _delta_log/_sidecar.

Include tests.

Additional context

No response

[scovich]

Wouldn't this be as simple as unconditionally calling Url::join, with the sidecar directory as the base and the sidecar.path as the input?

An absolute URL (with a scheme) as input replaces the whole base URL (even the scheme).

(seems like we anyway have to do that at some point, if we actually want to read the sidecar file??)

[OussamaSaoudi]

@scovich nice find! I was just able to verify it:

    let base = Url::parse("https://bucket-name.com/path/to/file/").unwrap();
    let url = "https://bucket-name.com/path/to/file/_sidecar/14567842.parquet";
    println!("{:?}", base.join(url).unwrap().to_string());

prints

https://bucket-name.com/path/to/file/_sidecar/14567842.parquet