refactor(rest): localize CreateTableRequest's partition-spec skip to REST

[rahulsmahadev]

Which issue does this PR close?

Follow-up to #2610. Addresses two review threads on that PR:

• @laskoviymishka: move the #[serde(skip_serializing_if = "Option::is_none")] off of the core UnboundPartitionField / UnboundPartitionSpec types and apply it only when these types are serialized as part of CreateTableRequest.
• @viirya (non-blocking): add a positive serialization test for UnboundPartitionField::field_id (set case).

What changes are included in this PR?

#2610 added #[serde(skip_serializing_if = "Option::is_none")] to UnboundPartitionField::field_id and UnboundPartitionSpec::spec_id so that REST CreateTableRequest payloads would omit unset partition ids. The same core types are also serialized as part of TableUpdate::AddSpec on the commit path, where a missing spec-id would hard-fail a Java REST server (PartitionSpecParser.fromJson throws IllegalArgumentException: Cannot parse missing int: spec-id). The breakage is latent today only because TableMetadataBuilder::add_partition_spec always populates spec_id via with_spec_id before pushing the AddSpec change.

This PR moves the skip behavior out of the core spec types and into the REST layer:

1. Drop #[serde(skip_serializing_if = "Option::is_none")] from UnboundPartitionField::field_id and UnboundPartitionSpec::spec_id. The core types again emit spec-id / field-id as null when unset, matching the pre-fix(rest): skip serializing unset optional fields in CreateTableRequest #2610 wire format and preserving the TableUpdate::AddSpec contract.
2. Add a serialize_with helper on CreateTableRequest::partition_spec that serializes through private shadow types whose fields skip the ids when None. This preserves the externally visible REST payload shape (partition-spec omitted entirely when unset; inner spec-id / field-id omitted when unset) without coupling the core spec type to it.

The shadow-type approach was chosen over a public newtype wrapper to keep the change scope to the REST crate and avoid touching iceberg::spec::partition's public surface beyond the revert.

Are these changes tested?

Yes. Four new / updated tests:

• crates/iceberg/src/catalog/mod.rs::test_add_spec_emits_spec_id_key: regression guard that TableUpdate::AddSpec always emits the spec-id key, including when spec_id is None. This is the test that would have caught the latent bug if fix(rest): skip serializing unset optional fields in CreateTableRequest #2610 had been merged without the add_partition_spec always-populates-spec_id invariant.
• crates/iceberg/src/spec/partition.rs::test_unbound_partition_field_field_id_set_is_serialized: positive test that a set field_id round-trips under the kebab-case field-id key (viirya's follow-up ask).
• crates/catalog/rest/src/types.rs::test_create_table_request_partition_spec_skips_inner_none_ids: verifies that the REST shadow serializer omits inner spec-id and field-id when unset, and emits them when set.
• crates/iceberg/src/spec/partition.rs::test_unbound_partition_spec_serialization_emits_none_as_null replaces the previous skip-expectation test and documents the new contract on the core type.

All pre-existing tests in iceberg (1361) and iceberg-catalog-rest (45) still pass. cargo fmt --check and cargo clippy --all-targets --all-features -- -D warnings are clean on both crates.

Test plan

• cargo test -p iceberg --lib
• cargo test -p iceberg-catalog-rest --lib
• cargo fmt --check -p iceberg -p iceberg-catalog-rest
• cargo clippy -p iceberg -p iceberg-catalog-rest --all-targets --all-features -- -D warnings