Gate Zed and Rider coverage; replace the Zed tests that asserted nothing

[MelbourneDeveloper]

.config/coverage/thresholds.json gated five packages — sharplsp, vscode-extension, and the three sidecars. Neither editor integration was one of them.

What was unguarded

Zed shipped 23 unit tests that were never executed. make _lint-zed runs clippy --all-targets, which compiles them; make test had no Zed target. Unrun, they had rotted:

Test	What it actually asserted
expected_version_matches_cargo_toml	EXPECTED_VERSION == env!("CARGO_PKG_VERSION") — and EXPECTED_VERSION is that env. Tautology.
server_binary_name_is_sharplsp	a const equals its own literal
expected_version_matches_extension_toml_version	that the crate version is non-empty. Its doc comment claims it proves extension.toml/Cargo.toml parity; it never opens extension.toml.
missing_binary_error_...	rebuilt the error string inside the test and asserted on its own copy — changing the real message in resolve_binary could not fail it

Rider had 2961 LOC of Kotlin, zero tests, and zero CI presence — not built, not linted, not tested on any PR. build.gradle.kts declared junit-jupiter, testFramework(TestFrameworkType.Platform), and tasks.test { useJUnitPlatform() } against no src/test directory at all.

What this adds

Package	Tests	Coverage	Gate
sharplsp-zed	23 → 31	75.41% → 85.04%	make _test-zed
sharplsp-rider	0 → 8	0% → 4.09%	make _test-rider (Kover)

Zed — the parse → read → enrich → format pipeline moves out of the zed::Worktree shell into pipeline.rs (98.41%) so it is reachable from a test; cached_or, with_default_log_level and tree_output split out of the WASM API adapters. The version-parity test now actually parses extension.toml, and the missing-binary test asserts on the message cached_or really produces.

Rider — first tests cover NuGetState: the installed-list merge, case-insensitive id matching (the two LSP responses do not agree on casing), sort order, and pending flags. Kover is wired to the repo ratchet at its honest 4.09%.

Two build defects had to be fixed before any Rider test could run:

• The platform test framework registers com.intellij.tests.JUnit5TestSessionListener, whose constructor loads junit.framework.TestCase. Without JUnit 4 on the runtime classpath it fails to instantiate and the task dies before a single test runs — NoClassDefFoundError, not a test failure.
• That framework also puts junit-vintage on the classpath, built against a newer junit-platform-commons than junit-jupiter 5.11.3 ships (support.scanning.ClassFilter). Vintage throws during discovery, aborting the whole task. Discovery is now Jupiter-only.

failOnNoDiscoveredTests = true stops an empty test task reporting BUILD SUCCESSFUL — which is exactly how this harness stayed green while containing nothing.

CI

New ci-editors.yml runs both legs. RIDER_REQUIRED=1 turns "no JDK 21+ found" from a local convenience skip into a hard CI failure, so the gate cannot silently skip. The Rider job also builds the plugin — it was never compiled on a PR before, so a Kotlin break could reach main behind a fully green pipeline.

Also

• default_threshold: 90 removed. check-coverage.mjs reads thresholds[project].line_percent and hard-fails on an unknown key; it never consulted default_threshold. Hard-failing on an ungated package is the correct strict behaviour and is what surfaced this gap.
• The JDK-21 discovery loop was duplicated per Rider make target — now once, in tools/rider/gradle.sh.

Verified locally

[sharplsp-zed]   coverage: 85.0394% (threshold: 85.04%)   31 passed, 0 failed
[sharplsp-rider] coverage: 4.0919%  (threshold: 4.09%)     8 passed, 0 skipped

make _lint-zed clean. make _build-rider still produces dist/sharplsp-rider.zip through the shared script.

Remaining debt

Tracked in #210: 95.9% of the Rider plugin is still untested (LspBridge, the tree nodes, PackageCardRenderer, PackageDetailsPanel, settings persistence — most needs BasePlatformTestCase fixtures), and Zed's 85% is near its ceiling as structured, since the remainder only exists inside Zed's WASM host.