Escape database, table and udf names for S3 paths

[kirillgarbar]

Unescaped paths may cause some errors with s3 client.
Check if escaped path exists is added for backward compatibility with old backups. path_exists was working only for files, so I added another method for directories.

[kirillgarbar]

Maybe it would be simpler to just try to use escaped path and retry with not escaped in case file is not found

[aalexfvk]

Let's do this way. It seems it is safe to escape the whole path at once.

[kirillgarbar]

Now I think adding retries to several methods will cause unnecessary complications of the old code. There are also some minor issues which will need to be addressed like list_dir used in check_data_part returning empty list even in case directory is not existing or the need to recollect all part paths in delete_data_parts in case of retry.

[aalexfvk]

How about do in such a way instead of passing a callable and kwargs. For me it is more clear.

    # In downloading methods
    remote_path = _udf_data_path(backup_meta.path, filename, escape=False)  # For backward compatibility
    remote_path = _get_escaped_if_exists(remote_path)

    def _get_escaped_if_exists(self, unescaped_path: str) -> str:
        """
        Return escaped path if it exists. Otherwise return regular path.
        """
        escaped_path = escape(unescaped_path)
        if self._storage_loader.path_exists(escaped_path, is_dir=True):
            return escaped_path
        return unescaped_path

[kirillgarbar]

Escaping the whole path looks like this ch_backup%2F7c543608%2D1af5%2D4eaa%2D9974%2D93b58e3f699c%2Fdata%2Funusual_char%2Ftab%07l%2E%2F%2D_e%2Fall_1_1_0 - '/' and '-' are escaped. We can ignore '/', but '-' was escaped in table names for things like metadata before, so ignoring it can lead to some troubles

[aalexfvk]

I see, we use safe="" for urllib.parse.quote. The escaping the whole string is a bad idea.
Ok let's return to the previous approach, it is more correct. But just rename function to _get_escaped_if_exists