[ZEPPELIN-6675] Prove the notebook route can consume the Shared Notebook Core port

[kimyenac]

What is this PR for?

Add a browser proof that the current Angular notebook route can hand one host-owned NotebookCorePort to a separately built React consumer. This builds on the ZEPPELIN-6674 port identity proof and moves it onto the real notebook route.

• The proof harness bootstraps the production WorkspaceModule and NotebookModule lazy routes and asserts that the activated component is the production NotebookComponent.
• It reads the activated /notebook/:noteId and /notebook/:noteId/revision/:revisionId parameters into one host-owned test Core. The React consumer reads the snapshot and receives route-driven subscription updates (note → other note → revision), always through the same port object.
• A browser-only MessageService double records the production component's getNote, noteRevision and listRevisionHistory requests, and the test asserts their exact order.

This is a seam-only proof. It does not switch the production renderer, move notebook state out of Angular, or implement re-subscription, recovery or mutation reducers (ZEPPELIN-6687).

The implementation is based on the ZEPPELIN-6675-notebook-route-boundary branch in the voidmatcha fork (823183338), which was written before ZEPPELIN-6674 merged. Porting it onto current master:

• CI wiring: the route proof runs in the integration-test phase behind web.e2e.core.port.proof.disabled, the same as the merged port identity proof, so it runs on the anonymous leg of run-playwright-e2e-tests. The branch bound it to the test phase with ${skipTests}, which frontend.yml's -DskipTests build skips.
• Shared static server: it now returns 404 for unknown paths, matching the merged ZEPPELIN-6674 review change, instead of serving index.html.
• Message double: it gains receiveEnvelope, added to MessageService by ZEPPELIN-6683. Without it, the workspace route did not render.
• Theme double: it gains getCurrentTheme and theme$. Without them, ThemeToggleComponent throws in ngOnInit. The original test did not catch this, because Angular's ErrorHandler reports it through console.error('ERROR', ...) and never as a pageerror. The test now fails on those errors too. Removing the theme fix makes it fail with this.themeService.getCurrentTheme is not a function.
• Lifecycle boundary is now asserted, not only documented:
  • the port's own keys are exactly getSnapshot and subscribe, and the port is frozen;
  • route activation and port consumption make no bootstrap, connect or close call.
• README: the section moves to the end of e2e/core-contract/README.md instead of splitting the replay subsections. It describes the actual CI wiring and notes that ZEPPELIN-6683's stale-reply rejection lives inside NotebookComponent and is not moved into the Core.
• Kept as is: the route path constants (notebook-route-boundary.ts). Production code changes are limited to extracting the existing route paths into those constants, which both the routing modules and the proof use.
• Dropped from the branch: the ./NotebookRouteBoundaryProbe webpack alias (it re-exposed the same module), and unused route-proof proofs state.

What type of PR is it?

Improvement

Todos

• Run the route proof on the production notebook routes with one host-owned port
• Assert route-driven snapshot and subscription updates through the same port
• Assert the port shape and that the connection lifecycle is not reached
• Run it through the normal browser CI path
• Record host-side responsibilities and what remains outside this proof

What is the Jira issue?

ZEPPELIN-6675

How should this be tested?

cd zeppelin-web-angular
npm run build:projects
npm run build:notebook-core-port-proof
npm run test:notebook-core-port-identity
npm run test:notebook-route-boundary

What I ran locally (after rebasing onto 6a5ed4459):

• Both browser proofs passed. Also passing:
  • test:notebook-core, typecheck:notebook-core
  • check:core-contract-fixtures, test:shell
  • prettier and eslint on the changed files
• ./mvnw verify -pl zeppelin-web-angular -Pweb-e2e (before the rebase): the proof build, the port identity proof and the route proof passed in integration-test.
• The main Playwright suite finished with 749 passed and 20 failed. Rerunning only the failing specs with one worker left 6 failures:
  • paragraph-functionality.spec.ts:174 and :335, on all three browsers;
  • both execute %python, and every failure context shows Fail to launch python process from my local Python setup;
  • the other 14 failures passed when run serially.
• Not run locally: the classic UI e2e. The maven run stopped after the main suite because I ran it without CI=true, so the HTML reporter waited for input. This PR does not touch zeppelin-web.
• I checked that each build setting kept from the branch is still required by building without it:
  • mathjax types: without them, the MathJax directive fails to compile;
  • the custom webpack builder: without it, Monaco CSS fails to parse;
  • the style include paths: without them, the components' .less imports fail to resolve.

Screenshots (if appropriate)

N/A.

Questions:

• Does the license files need to update? No. New files carry the ASF header.
• Is there breaking changes for older versions? No. Route shapes and behaviour are unchanged.
• Does this needs documentation? e2e/core-contract/README.md is updated.

Unrelated, found while testing: production main.js builds the Monaco codicon.ttf URL from the build machine's absolute file:/// path, so the browser refuses to load that font. It also happens on master.

🤖 Generated with Claude Code

[voidmatcha]

If Chromium fails to launch, the harness never returns its cleanup handle, leaving the HTTP server open. We close local test servers on failure in capture-fixtures.spec.ts. Could we cover this failure path here too?

Suggested change

	const browser = await chromium.launch();
	let browser;
	try {
	browser = await chromium.launch();
	} catch (error) {
	await new Promise(resolve => server.close(resolve));
	throw error;
	}

[kimyenac]

Applied in f38120b. I named the callback resolveClose instead of resolve, because this file imports resolve from node:path. I checked the failure path by pointing PLAYWRIGHT_BROWSERS_PATH at an empty directory. Before the change the process never exited because the server stayed open. After it, the proof fails immediately with the launch error.