Type system for pipelined SQL

[mav3ri3k]

Prequel stand for Pipelined Relational Query Language.
Fair to say the killer feature for me while using PRQL has been pipelining.
Nushell also functions in quite same regard where everything can be table which can be filtered through pipes.

It has way to create tables and a type system for passing data through pipes:

# nushell

# describe input through a pipe
ls | describe
table<name: string, type: string, size: filesize, modified: date> (stream)

and query over data passed through pipes

ls | get name | filter { $in =~ "Cargo" } | describe
list<string> (stream)

The PRQL book references Type System.
(Excellent read for me. This could also be linked under roadmap on the website.)
I am particularly excited about it. This would be excellent for lsp.
I think some implementation details can be referenced from nushell.
Nu is interpreted and the type check happens as the query is being run through pipes.
PRQL would likely not be able to do a complete type check over a query during compile time.
For example in above case, the filter may return as none, but that can not be validated during compile.

I have few questions too:

1. Schema Definition Language

Once a Type System is worked upon, what would be scope for implementing a SDL ?
Some solutions in similar respect already exist.
The project has often mentioned that it is focused on querying side.
However it is not clear to me if having a SDL would be too much overhead after having a Type System.

2. Physical Layout

The Type System page also has section for Physical layout.
Having Apache Arrow as common physical layout would be excellent. However, I am really confused here.
PRQL is a language which can map to other query language ( or maybe also directly to query engine like Datafusion ? ).
What does "physical layout" mean in this context where PRQL is more like a modern abstraction for querying ?

3. Can PRQL be directly mapped to query engines like Datafusion, Polars, etc ?

The title.