feat(cpp): add C++ packaging for protobuf, ANTLR and extensions

[nielspardon]

Adds C++ support to the packaging machinery, alongside Python, Rust and Go. Based on main independently of the Go PR (#32) so it can be reviewed/merged on its own.

What this adds

Three CMake source packages under cpp/, distributed as git tags consumed via CMake FetchContent (tag-as-package model, like Go — no registry, no credentials; pushing cpp/<pkg>/vx.y.z publishes it):

• substrait-protobuf — vendors the .proto files and generates C++ at build time with the consumer's own protoc via protobuf_generate. Generated .pb.cc is deliberately not committed: C++ protobuf code is ABI-coupled to a protobuf runtime, so committing it would pin every consumer to one protobuf version. Exposes substrait::proto + a find_package(SubstraitProtobuf) config. This matches how substrait-cpp consumes the protos today.
• substrait-antlr — commits parsers from the stock ANTLR C++ target (no fork, unlike Rust) and builds the ANTLR C++ runtime hermetically via FetchContent (pinned to the generator version), so no system/conda ANTLR runtime is required on any platform. Exposes substrait::antlr.
• substrait-extensions — data-only: C++ has no canonical YAML/JSON schema code generation (no typify / datamodel-code-generator equivalent), so it vendors+installs the raw spec data and exposes SubstraitExtensions_DATA_DIR; typed parsing lives downstream.

Generated/vendored content lives only on tags; main keeps scaffolding (CMakeLists, generate scripts, READMEs, cmake config templates, smoke tests, .gitkeep placeholders).

Tooling & CI

• pixi.toml: cpp-generate-{protobuf,antlr,extensions} tasks + cmake/cxx-compiler/protobuf deps.
• scripts/cpp/next_alpha_version.sh (alpha-by-default, mirrors Go); reuses tag_exists.sh and attach_subtree.sh.
• Workflows: cpp_publish.yml → cpp_{protobuf,antlr,extensions}.yml (generate → cmake build → ctest → commit → tag-push), ci_cpp.yml (validates on cpp/** PRs), and a publish-cpp-artifacts job in publish_artifacts.yml.
• README updated with the C++ workflow hierarchy and a "C++ Code Generation" section.

Consumer adoption

Intended to let substrait-cpp and the DuckDB Substrait extension eventually drop their third_party/substrait submodule + bespoke protoc machinery for a pinned FetchContent dependency. Adoption is a follow-up conversation with the substrait-cpp maintainers; this PR just produces the packages.

Verification

Each package builds + passes a ctest smoke test (compile a generated proto header / lex+parse a type / resolve an extension data file). Verified locally end-to-end (pixi run cpp-generate-* → cmake build → ctest) and a protobuf cmake --install produces the full find_package surface. ci_cpp.yml runs the same generate+build+test against the latest spec release on every cpp/** PR.

🤖 Generated with AI