[Improvement] Infer Data Lakes schemas by reading the metadata

[alifirat]

Use case

As a User, I would like to have an instant creation when creating a DataLake table like Iceberg or Delta Lake.

Describe the solution you'd like

Today, I have the feeling that the schema inference of the Data Lakes is inferred by reading the data files directly. The main issue I see with this approach is that if we have millions of data files, it's going to take a while to just infer the schema.

To me, the schema can be infer by reading the metadata:

• For Iceberg, if you reads the v{N}.metadata.json there is a field called schemas that contains all the fields.
• For Delta Lake, if you the reads the path metadata.schemaString you also have the schema.

To me, the improvements are going to be on 2 side:

• performance: reading a simple JSON file on S3 against reading N files on S3
• cost: 1 single AWS call for the schema vs multiple calles

A potential drawback if the schema changes during the time:

• Well in this case, as like any ClickHouse table today, the user may need to run an alter table command or recreate the table since it's a read-only mode ?

Describe alternatives you've considered

Don't look to alternatives.

Additional context

See #49958 #49959

[kssenii]

For Iceberg we do already infer schema from metadata (from #55695), for DeltaLake it will be added once #63201 is merged.