[Peefy]

Sits directly on main — #399 (the v0.13.1 release bump) is merged, so this PR's diff is exactly the three commits below.

Follow-up to #399. That PR replaced the checked-in binaries with the v0.13.1 release and fixed the two CI failures the bump caused; this one removes the workarounds that release made dead, and corrects the docs that described the old behaviour.

Why the guards are dead

v0.13.1 closes three RPC-registry gaps the bindings had been compensating for:

• BuiltinService.ListMethod and BuiltinService.Ping are registered. Bindings that treated an empty method table as "this RPC doesn't exist" now always get a table.
• KclService.FormatTestReport and the five Generate* RPCs (GenerateToml, GenerateKcl, GenerateOpenAPI, GenerateProto, GenerateDoc) are registered, so the skip guards in front of their assertions never fire.
• KclService.ListDepFiles was removed, replaced by new LoadPackageResult.imports / kcl_mod / apps fields.

28 RPCs are now advertised: 26 under service KclService plus Ping and ListMethod under service BuiltinService.

What this changes

Removes the t.Skip / pytest.skip / markTestSkipped / pending() / empty-table guards across go, python, dart, zig, julia, lua, nodejs, java, kotlin, dotnet, ruby, swift, cpp and replaces them with the assertions they were written to enable.

Where a guard was the only thing keeping a test green, the assertion now pins the registry size at 28 — so the next core-side RPC removal cannot pass unnoticed. ListDepFiles is exactly the case that slipped through: it was deleted from the core and these bindings kept carrying a skip for it.

Doc corrections that the release made stale:

• dart/lib/src/kcl_lib.dart — the old comment claimed spec.proto declares FormatCode / LintPath / etc. under BuiltinService. It does not; the core registers every RPC under KclService except the two spec.proto puts in BuiltinService.
• zig/README.md RPC count (20 → 26), julia/README.md, dart/README.md, wasm/README.md getVersion examples
• docs/architecture.md — the ListDepFiles drift entry is now closed, and item A's core half landed in v0.13.1

Deliberate non-changes: the per-case new_core skips (python, ruby, swift, cpp, lua, dotnet, kotlin, java, go, c) stay. They query the live core through list_method() and skip only when the loaded core genuinely does not advertise the RPC — an evidence-driven capability check, not a v0.13.0 artifact.

cpp: three separate things

The six examples asserted nothing. The Generate* / FormatTestReport examples each carried a comment saying the linked runtime "currently prints nothing" — and then asserted nothing at all, so cpp-test's exit 1-on-failure loop had nothing to catch. They all produce real output on 0.13.1; they now check it.

cpp/src/lib.rs hand-copied 12 proto messages for these RPCs, on the stated grounds that "the kcl-api revision pinned in Cargo.lock predates the RPC, so the wrapper has no generated counterpart". It does have them: kcl_api::FormatTestReportArgs, GenerateTomlArgs, GenerateKclArgs, GenerateOpenApiArgs, GenerateProtoArgs, GenerateDocArgs (and the matching *Result) are all generated at the pinned rev, field-for-field. The copies are gone. The From impls stay — they convert the cxx types to the prost ones — and so do call_rpc and list_method, because kcl_api still exposes no typed client method for these, only kcl_api::call.

The first CI run of this PR caught a build break in the above. call_rpc takes &Req with Req: prost::Message, so call_rpc(b"...", &args.into()) leaves Req pinned only by the argument; Into has too many candidate impls and the type parameter is never pinned elsewhere, so all six dispatch functions fail with E0283. Fixed by binding the conversion to a named local — which also documents which generated message each RPC expects. Worth noting how this survived local testing: cpp/CMakeLists.txt's add_custom_command lists ${RUST_BRIDGE_CPP} ${RUST_LIB} ${RUST_HEADER_FILE} as outputs but has no DEPENDS on ${RUST_SOURCE_FILE}, so editing cpp/src/lib.rs does not re-run cargo build and the previous ctest/consistency run was against a stale staticlib. That is a pre-existing bug in the build, left alone here — the re-verification below builds the crate explicitly first.

The one skip that was not a version artifact

swift/Tests/KclLibTests/AstJsonAlignmentTest.swift skipped on "the pinned kcl runtime does not copy the parent into mixins". That is not what the pin does. The parser fills mixins only from an explicit mixin [...] clause (crates/parser/src/parser/stmt.rs); the parent goes to parent_name alone. No future pin will make that skip stop firing, so it is replaced with the invariant it was standing in for — a schema with a parent and no mixins has an empty mixins.

The stale ruby/Cargo.lock is not here — it is release correctness, so it rides in #399 with the manifest bump.

Verification

Run locally against the v0.13.1 darwin-arm64 dylib:

suite	result
swift	102 tests, 0 failures, 0 skipped
ruby	145 runs, 654 assertions, 0 failures, 0 errors, 0 skips
cpp	cargo build --release clean · 9/9 ctest · 13/13 consistency cases, 0 skipped · all 31 examples exit 0 under the CI loop · cargo fmt --check adds no new diffs (3 pre-existing at 1775/1805/2328)
go (native/path/plugin)	ok — the 6 Generate* / FormatTestReport guards now assert instead of skipping
python	160 passed, 2 skipped (skips want PyYAML, unrelated)
nodejs	109 passed (listMethod returns 28)
dart	124 passed, dart analyze clean
zig	exit 0 (BuiltinService.Ping now dispatches)
julia	129 + 106 + 168 + 18 (coverage and test report are populated)

swift, ruby and cpp all report zero skips, which is the direct evidence that every guard removed here was dead.