Support namespace for Tensorflow

[yuqi1129]

As Lance-ray and Lance-spark have supported namespace to manage tables, we also hope Tensorflow can use namespace to organize tables.

[westonpace]

Does tensorflow do table management? I did not think it was in the same category as spark and ray. What does it mean for tensorflow to organize tables?

[yuqi1129]

I mean, we can use namespace to organize the table, and we will not use the table path directly, as lance-ray and lance-spark do.

The following is an example for lance-ray, and I would like to make tensorflow support namespaces too.

ray.init()

# Connect to a metadata catalog (directory-based example)
namespace = ln.connect("dir", {"root": "/tmp/lance"})

# Create a Ray dataset
data = ray.data.range(1000).map(lambda row: {"id": row["id"], "value": row["id"] * 2})

# Write to Lance format using metadata catalog
write_lance(data, namespace=namespace, table_id=["lance_catalog", "schema", "my_table"])

Does tensorflow do table management?

Of course, tensorflow can't manage tables, maybe my expression caused you to misunderstand.

[westonpace]

Ok, I understand now. The goal is for tensorflow to read tables (which tensorflow does need to be able to do) but to do so using the namespace API instead of requiring dataset URIs.

Would this code go in this repository? Is this something that could be placed in lancedb? Or would you need to create a new subproject?