Bundle Monaco through the Angular build (ESM)

[nabramovitz]

Closes #5561.

What changes

Monaco previously loaded through its own AMD loader from a wholesale copy of the vs/ distribution in assets: unversioned URLs that can never be served cache-control: immutable (one of the two cache-policy blockers the issue records), a serialized loader → editor.main.js → css sequence, and a peer conflict — @cfstratos/monaco-yaml 2.5.0 pins monaco ^0.20 while the repo ships 0.52, which breaks plain npm install.

• The editor is now a hashed lazy chunk (673 kB wire, vs 966 kB across three unversioned requests), dynamically imported on first editor use; the copied vs/ asset tree is retired. Initial bundle is unchanged — Monaco stays off the critical path (perf(core): lazy-load monaco, right-size assets; add load diagnostics #5560's lazy-load behavior preserved).
• Language workers are bundled from relative new Worker(new URL(...)) references via one-line wrapper modules (the builder doesn't resolve bare package specifiers there). Only the languages Stratos actually edits get a worker — json and yaml; every editor surface in the repo is json/yaml/plaintext (verified by census of all app-monaco-editor usages and createModel/setModelLanguage call sites). Other labels fall back to the basic editor worker: syntax highlighting is main-thread and unaffected, workers only add validation/completion.
• @cfstratos/monaco-yaml → upstream monaco-yaml 5.5.1 (peers monaco-editor >= 0.36, actively maintained). Registered once by the ESM loader; schema registration goes through a configureYaml() helper wrapping the v5 update() handle — same global last-writer semantics as the old yamlDefaults surface. The Helm chart-values editor drops its AMD contribution load and now surfaces schema-registration failures instead of losing them as unhandled rejections.
• Two toolchain notes baked into the diff: monaco's codicon font needs a .ttf file-loader entry in angular.json, and the dynamic import uses the concrete esm/vs/editor/editor.main.js path because the package publishes only a module field — esbuild accepts the bare specifier but vitest's vite resolver does not (typed by a small local ambient declaration).

Verification

• Full gate green (lint, all unit tests, production build) on the final tree.
• Monaco had zero e2e coverage; this adds a runtime smoke that drives the variable-edit dialog on a real app: editor renders from the lazily imported chunk, a keystroke round-trips, and an invalid yaml model produces diagnostics from the monaco-yaml worker. Green through the live harness.
• Two adversarial self-review passes; the notable catch was the chart-values editor's init path still calling the removed AMD window.require — exactly the k8s surface the CF-side smoke doesn't reach.

Follow-on candidate (noted, not done here): import a tree-shaken feature subset via editor.api instead of full editor.main to shrink the editor chunk further.

Net −231 lines.

[norman-abramovitz]

LGTM