Add the core-jvm-gradle-plugin artifact of CoreJvm Compiler

[alexander-yevsyukov]

Since core-jvm-compiler 2.0.0-SNAPSHOT.090, the CoreJvm Gradle Plugin lives
in its own thin JAR artifact, core-jvm-gradle-plugin, and core-jvm-plugins
carries only the Compiler plugins. The POM of the plugin artifact declares a
runtime dependency on the latter, so consumers receive the same graph as
before the split.

CoreJvmCompiler speaks in terms of roles

Both version and dogfoodingVersion move to 2.0.0-SNAPSHOT.090, the first
release providing the new artifact. With no supported version predating the
split, the properties naming retired artifacts or packaging details are
removed rather than deprecated:

• pluginLib (property and function) named an artifact that no longer exists;
• the fatJar terminology exposed an implementation detail that may change
  without affecting consumers.

The object now names roles instead of packaging:

Property / function	Resolves with
gradlePlugin, gradlePlugin(version)	dogfoodingVersion — the build classpath
compilerPlugins, compilerPlugins(version)	version — integration tests
gradlePluginArtifact, compilerPluginsArtifact	artifact names

Consumers referencing pluginLib or pluginsArtifact must move to the new
names on their next ./config/pull.

See SpineEventEngine/core-jvm-compiler#111 for the module split.

.idea/kotlinc.xml becomes project-local

IDEA rewrites this file on its own — Kotlin JVM target and bundled plugin
version — so a tracked copy churns on every machine. It is now ignored and
untracked here, and consumers get the same treatment on their next
./config/pull:

• migrate generalizes the misc.xml handling into a loop over both
  IDE-managed files, covering the untracking and the preserve/restore
  around the .idea overlay, so a consumer's own Kotlin settings survive a
  pull instead of being clobbered by whatever the puller has locally.
• update-gitignore.sh retires the !.idea/kotlinc.xml negation the baseline
  shipped from 2021 until 2025. A legacy raw-copied .gitignore still carries
  it, and — .gitignore being last-match-wins — it re-includes the file the
  current baseline ignores.
• Adding a second entry to retired_negations surfaced a latent bug: awk -v
  applies escape processing and cannot carry an embedded newline, so the
  documented "newline-separated" list failed on its first multi-entry use.
  Both call sites now read it through ENVIRON[].

Tests

• scripts/test-update-gitignore.sh covers the new negation in both merge
  paths (steady-state and first-migration).
• scripts/test-migrate-ide-files.sh is new: it runs the real migrate
  against a throwaway consumer that starts with both files committed and both
  retired negations present, then asserts they end up untracked, on disk,
  ignored, carrying the consumer's own content, immune to git add -A, and
  that a second run is a quiet no-op leaving a clean tree.

Both suites were mutation-tested — each was confirmed to fail when the
mechanism it guards is deliberately broken.

Dependency bumps

Base → 2.0.0-SNAPSHOT.441, Compiler → 2.0.0-SNAPSHOT.067, ProtoTap
→ 0.17.0. Paired constants (versionForBuildScript, fallbackDfVersion)
move with their primaries; all sibling artifacts were verified published at
the target versions.

Task plan: maven-central-throttling

Adds .agents/tasks/maven-central-throttling.md, drafted in
core-jvm-compiler after the 2026-08-19/20 Maven Central consumption-limit
incident, with three decision records from its review here:

• Windows CI cache (deferred).windows-latest runs in exactly one
  workflow, triggered on pull_request only. GitHub caches cross branches
  only from the default branch, so no shared Windows entry can ever be
  written and every Windows PR job faces Central cold — precisely the
  observed 403 on kotlin-stdlib. Workflow changes are held; re-open on
  the next Central-caused CI failure.
• TestKit migration (set aside). It would move the traffic to the
  uncached .gradle-test-kit directory rather than remove it, and cost the
  white-box assertions these specs rely on.
• Rollout on hold. The root cause is now filed upstream as
  ProjectBuilder silently bypasses the dependency cache, causing full re-downloads per test run — now triggers Maven Central 429 blocks gradle/gradle#38915, with Reuse dependency caches in ProjectBuilder gradle/gradle#38918 fixing it. If that lands, the
  planned fixture rollout shrinks to a Gradle wrapper upgrade.

Note that .agents/tasks/ is repo-local by design — migrate and
adopt-shared-agents deliberately skip it — so this file records the plan in
config and does not propagate to consumers.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:a8f55a7e8a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".