Build & release plumbing: make the artifacts consumable

Goal: publish to Maven Central + PyPI.

Today no third party can depend on any part of this repo:

• No Maven reactor. 13 poms, zero <modules> blocks. Cross-module edges are hardcoded versions with no dependencyManagement, so several modules only build after a manual mvn install of their upstreams in the right order.
• The main module hard-depends on an unpublished artifact (org.jagentic:jagentic-core:0.1.0), so a clean checkout cannot run the documented mvn clean test.
• No release plumbing anywhere: no distributionManagement, sources/javadoc jars, GPG signing, or the POM metadata (url, licenses, developers, scm) that Central requires. 0 git tags.
• Namespaces are unverifiable: org.agentic.flink / org.jagentic imply domains the project does not own. Needs a groupId tied to an owned identity, and flink removed from the groupId.
• Go module path 404s (github.com/jagentic/goagentic); the importable path is github.com/Ugbot/Agentic-Streaming/ports/go.
• The pyagentic name is already taken on PyPI; agentic-flink has never published (404), and its wheel ships no jar so pip install alone is non-functional.
• Dependency hygiene: flink-cep leaks at compile scope, and consumers are forced to take a Postgres driver, HikariCP, four LLM SDKs and a log4j backend.

Good news: the thin-jar/-uber-jar split is already correct and should be left alone.

Roadmap epic. Detailed findings, file:line citations and sub-tasks live in the AI-grind tracker (project AGS), which is the source of truth for this work. Local dev — build, test, tracking, profiling — is managed with AI-grind.

Tracked as AGS-1 in the devtools-mcp tracker.

Start with the 13 Maven POMs and the main module dependency, then inspect ports/go and the package layout. Use the AI-grind AGS tracker as the source of truth and run the documented mvn clean test workflow. Done means the modules build from a clean checkout and the Java, Go, and Python artifacts are consumable through their intended registries.