feat(python): support DataFrame init from pydantic model data

[alexander-beedie]

Closes #8154.

This was a simple addition; pydantic model data is largely equivalent to @dataclass or NamedTuple data, which we already support. Only needed a few minor tweaks to leverage the existing code.

• No additional dependencies (except for unit tests; integration is optional/lazy).
• Works with instantiated sequences (list/tuple) of models and/or generators of such.

Example

from pydantic import BaseModel
import polars as pl

class Shipment( BaseModel ):
    exporter: str
    importer: str
    tonnes: int
    
shipments = [
    Shipment( exporter="Sri Lanka", importer="USA", tonnes=10 ),
    Shipment( exporter="India", importer="UK", tonnes=25 ),
    Shipment( exporter="China", importer="UK", tonnes=40 ),
]

pl.DataFrame( shipments )

# shape: (3, 3)
# ┌───────────┬──────────┬────────┐
# │ exporter  ┆ importer ┆ tonnes │
# │ ---       ┆ ---      ┆ ---    │
# │ str       ┆ str      ┆ i64    │
# ╞═══════════╪══════════╪════════╡
# │ Sri Lanka ┆ USA      ┆ 10     │
# │ India     ┆ UK       ┆ 25     │
# │ China     ┆ UK       ┆ 40     │
# └───────────┴──────────┴────────┘

[stinodego]

Great! My colleague will be very happy with this 😄

[ritchie46]

@samuelcolvin FYI. :)

[samuelcolvin]

amazing!

As it happened I've started using pola.rs today, and it's been an incredible experience - I'm really impressed. Thank you.

[alexander-beedie]

As it happened I've started using pola.rs today

@samuelcolvin: Well now you've made me look into the pydantic source to work out what the optimal loading path is and make sure we're doing this as fast as possible: #8181 🤣