How to get polars.LazyFrame.pivot work with multiple columns?

The typehint also shows a DataFrame being accepted: oncolumns: Sequence[Any] | Series | DataFrame

oncolumns = pl.DataFrame(
    [["maths", "theory"], ["maths", "practice"], ["physics", "theory"]],
    orient="row",
    schema=["subject", "exam"]
)

# shape: (3, 2)
┌─────────┬──────────┐
│ subject ┆ exam     │
│ ---     ┆ ---      │
│ str     ┆ str      │
╞═════════╪══════════╡
│ maths   ┆ theory   │
│ maths   ┆ practice │
│ physics ┆ theory   │
└─────────┴──────────┘

lf.pivot(on=["subject", "exam"], index="id", oncolumns=oncolumns).collect()

# shape: (2, 4)
┌─────┬────────────────────┬──────────────────────┬──────────────────────┐
│ id  ┆ {"maths","theory"} ┆ {"maths","practice"} ┆ {"physics","theory"} │
│ --- ┆ ---                ┆ ---                  ┆ ---                  │
│ i64 ┆ i64                ┆ i64                  ┆ i64                  │
╞═════╪════════════════════╪══════════════════════╪══════════════════════╡
│ 3   ┆ null               ┆ null                 ┆ 15                   │
│ 1   ┆ 10                 ┆ 12                   ┆ null                 │
└─────┴────────────────────┴──────────────────────┴──────────────────────┘

Or in the case of your Series example .collect().to_series() would be .unnest("subject").collect() instead.