[All SDKs] Add dynamic proto descriptor generation from UC schema

[elenagaljak-db]

Summary

Recently there was a PR which added a schema module to the Rust SDK (PR #200) that builds a protobuf
DescriptorProto at runtime from a Unity Catalog table schema, eliminating the need for offline .proto file generation.

We should port this capability to all other SDK languages:

• Python
• TypeScript
• Java
• Go

Motivation

Today, SDK users must run generate_proto as a build step before they can stream data. With dynamic descriptor generation, a caller can query the UC REST API for table metadata and immediately start streaming — no code-gen step required.

Behavior to Implement

Each language SDK should expose:

• A UcColumn type (mirrors UC REST API: name, type_name, type_text, type_json, nullable, position)
• A UcTableSchema type (schema_name + table_name + columns)
• descriptor_from_uc_columns(columns, message_name) → proto descriptor
• descriptor_from_uc_schema(schema) → proto descriptor

The type mapping and encoding contracts (e.g. DATE = int32 days-since-epoch, TIMESTAMP = int64 microseconds UTC) are documented in the Rust implementation: rust/sdk/src/schema.rs.

Reference Implementation

Rust SDK: rust/sdk/src/schema.rs (added in PR #200 / commit 1ebda6e)

Notes

• STRUCT, ARRAY, and MAP columns require type_json to be populated (the JSON from the UC REST API /api/2.1/unity-catalog/tables/{name} response)
• Simple scalar columns only need type_name
• Max nesting depth should be capped (Rust uses 100) to prevent stack overflow on pathological inputs