Support multi-argument transforms

### Feature Request / Improvement

Part of #1818.

The V3 spec generalizes partition and sort fields to multiple source columns via `source-ids`. PyIceberg currently raises when `source-ids` has more than one element, so loading such tables fails — but the spec requires v3 readers to read tables with unknown transforms, ignoring them (null partition values, always-true projection in scan planning).

Scope: model `source-ids`, serialize per spec (`source-id` for single-argument transforms, `source-ids` otherwise), and treat multi-argument transforms as unknown in scan planning. Evaluating multi-argument transforms on write is out of scope until the spec defines concrete ones.

Start by tracing the partition-field model and its serialization, then follow scan-planning handling for unknown transforms. Compare the current single-argument source-id behavior with the V3 specification. Done means multi-argument source-ids load and serialize correctly, are treated as unknown during scan planning, and write-time evaluation remains out of scope.

Written by the indexing model from the issue text.