feat(python): Support Series init as struct from @dataclass and annotated NamedTuple

[alexander-beedie]

Closes #5022

---

New functionality

• Extends existing polars DataFrame support for @dataclass and NamedTuple init to Series .
• Improves type-inference from annotations (specifically from ForwardRef and explicit Optional).

Setup (example)

from dataclasses import dataclass
from datetime import datetime

@dataclass
class Trade:
    timestamp: datetime
    ticker: str
    price: float
    size: int | None = None

trades = [
    Trade(datetime(2022, 10, 1, 14, 30, 45), "AAPL", 138.20, 250),
    Trade(datetime(2022, 10, 2, 10, 15, 12), "FSLY", 9.16, 2000),
    Trade(datetime(2022, 10, 3, 15, 30), "MU", 50.10, 600),
]

Initialise series

import polars as pl
s = pl.Series( trades )

Before (object)

# shape: (3,)
# Series: '' [o][object]
# [
#     Trade(timestamp=datetime.datetime(2022, 10, 1, 14, 30, 45), ticker='AAPL', price=138.2, size=250)
#     Trade(timestamp=datetime.datetime(2022, 10, 2, 10, 15, 12), ticker='FSLY', price=9.16, size=2000)
#     Trade(timestamp=datetime.datetime(2022, 10, 3, 15, 30), ticker='MU', price=50.1, size=600)
# ]

After (struct)

# shape: (3,)
# Series: '' [struct[4]]
# [
#     {2022-10-01 14:30:45,"AAPL",138.2,250}
#     {2022-10-02 10:15:12,"FSLY",9.16,2000}
#     {2022-10-03 15:30:00,"MU",50.1,600}
# ]

Suitable test coverage added for all the above.