fix(pivot-wider): handle the case of empty id_cols

[cpcloud]

Fixes an issue with handling simpler cases of Table.pivot_wider where you want to take two columns and use one as the header and one as the first and only row.

[cpcloud]

cc @IndexSeek

[IndexSeek]

Thank you for working on this! I wanted to add a bit of context to the problem I was seeing here and to help repro/explain the desired outcome.

Example data

from ibis.interactive import *

t = ibis.memtable(dict(outcome=["yes", "no"], counted=[3, 4]))
t

┏━━━━━━━━━┳━━━━━━━━━┓
┃ outcome ┃ counted ┃
┡━━━━━━━━━╇━━━━━━━━━┩
│ string  │ int64   │
├─────────┼─────────┤
│ yes     │       3 │
│ no      │       4 │
└─────────┴─────────┘

I was hoping to "transpose" this with pivot_wider.

Failed attempt

t.pivot_wider(names_from="outcome", values_from="counted")

Expected signature: GroupedTable(table: Relation, groupings: Annotated[tuple[Value, ...], Length(at_least=1, at_most=None)], orderings: tuple = (), havings: tuple = ())

Hacky workaround

t.mutate(dummy=1).pivot_wider(names_from="outcome", values_from="counted").drop("dummy")
┏━━━━━━━┳━━━━━━━┓
┃ yes   ┃ no    ┃
┡━━━━━━━╇━━━━━━━┩
│ int64 │ int64 │
├───────┼───────┤
│     3 │     4 │
└───────┴───────┘

Relatively complex workaround

cases = {
    outcome: ibis.case()
    .when(_.outcome == outcome, _.counted)
    .end()
    for outcome in ("yes", "no")
}
t.select(**cases).agg(s.across(s.all(), _.sum()))
┏━━━━━━━┳━━━━━━━┓
┃ yes   ┃ no    ┃
┡━━━━━━━╇━━━━━━━┩
│ int64 │ int64 │
├───────┼───────┤
│     3 │     4 │
└───────┴───────┘

[jcrist]

Overall LGTM. If you're adding none in #9917 I guess we'll want to hold-off merging this and rebase after that one.

[cpcloud]

Yep, SGTM, I started the refactor after this PR, so I'll merge that one first and then this one.

Thanks for the review(s)!