[gty404]

Changing only a field's documentation through SetCurrentSchema silently reuses the previous schema ID and discards the new doc. Include doc in SchemaField equality so schema deduplication records doc-only changes, including nested fields, while preserving the existing default-value comparisons.

Add regressions for required/optional fields, nested struct/list/map fields, doc replacement/addition/removal, metadata JSON round trips, identical-schema deduplication, and historical schema ID reuse. Update two metadata deserialization expectations to include the doc already present in their JSON fixtures.

Validation:

• The three new regression tests failed before the fix and passed afterward in the original checkout.
• Rebuilt and ran schema_test, table_test, and json_serde_test on the PR branch based on current upstream main.
• Ran clang-format and git diff --check.

[zhjwpku]

Good catch, seems like an oversight of #96

[shangxinli]

nit: doc_ is a plain std::string, so no-doc and doc set to empty compare equal here, while Java's Objects.equals at Types.java:983 treats null and empty as different schemas.

[gty404]

Thanks for pointing this out. The current C++ API already represents both an absent doc and an explicitly empty doc as an empty std::string. Deserialization defaults a missing doc to an empty string, and serialization omits empty docs, so that distinction is lost before equality is checked. This PR preserves that existing behavior while making changes to the doc string participate in schema equality. Matching Java’s null-versus-empty semantics would require a separate change to the doc representation, API, and serialization.