feat: Streamline use of predicates connected by & with IEJoin (join_where)

[alexander-beedie]

Following on from #19547.

Small ease of use improvement for predicate conjunctions passed to join_where.

Decomposes And conjunctions internally, rather than raise an unnecessary error (as it does now), so the caller can create and pass a single expression (instead of n expressions). Handles nested And (eg: p1 & ((p2 & p3) & p4))

Example

The following are all now allowed, and equivalent:

df1.join_where(
    df2,
    predicate1,
    predicate2,
    predicate3,
    predicate4,
)

df1.join_where(
    df2,
    predicate1 & predicate2,
    predicate3 & predicate4,
)

df1.join_where(
    df2,
    predicate1 & predicate2 & predicate3 & predicate4
)

[coastalwhite]

The plan originally was to integrate the join_where to use the same code as the collapse_joins optimization pass. That pass already does this. Maybe it is possible to reuse that code?

[ritchie46]

Yes, we need to merge those paths once. Will not block this until then.