CI: cleanup and fix broken job references

- Fix code-size -> code-analysis job name (5 jobs had wrong dependency)
- Remove useless install-check job (was no-op)
- Add explicit docs_only guard to release-check
- Remove dead submodule checkout steps (no submodules in repo)
- Rename detect-docs-only -> detect-docs-changes, add docs_changed output
- Reorder check script: format first for faster fail
- Fix billing error test (PR #12946 removed fallback detection but not test)

docs/ci.md

@@ -19,10 +19,10 @@ Tier 1 — Cheapest gates (parallel, ~43 s)
   check-format     secrets
 
 Tier 2 — After format (parallel, ~2 min)
-  check-lint       code-size
+  check-lint       code-analysis
 
 Tier 3 — Build (~3 min)
-  build-artifacts   install-check
+  build-artifacts   release-check
 
 Tier 4 — Tests (~5 min)
   checks (node tsgo / test / protocol, bun test)
@@ -39,8 +39,8 @@ Tier 5 — Platform (most expensive)
 ```
 docs-scope ──► changed-scope ──┐
                                 │
-check-format ──► check-lint  ──►├──► build-artifacts ──► checks-windows
-             ├─► code-size  ──►├──► install-check
+check-format ──► check-lint  ──►├──► build-artifacts ──► release-check
+             ├─► code-analysis ►│                    └──► checks-windows
                                 ├──► checks
                                 ├──► macos
                                 └──► android
@@ -65,37 +65,37 @@ secrets (independent)
 
 ### Tier 2 — After Format
 
-| Job          | Runner            | Depends on     | Purpose                                                     |
-| ------------ | ----------------- | -------------- | ----------------------------------------------------------- |
-| `check-lint` | Blacksmith 4 vCPU | `check-format` | Runs `pnpm lint` — cleaner output after format passes       |
-| `code-size`  | Blacksmith 4 vCPU | `check-format` | Checks LOC thresholds — accurate counts need formatted code |
+| Job             | Runner            | Depends on     | Purpose                                                     |
+| --------------- | ----------------- | -------------- | ----------------------------------------------------------- |
+| `check-lint`    | Blacksmith 4 vCPU | `check-format` | Runs `pnpm lint` — cleaner output after format passes       |
+| `code-analysis` | Blacksmith 4 vCPU | `check-format` | Checks LOC thresholds — accurate counts need formatted code |
 
 ### Tier 3 — Build
 
-| Job               | Runner            | Depends on                | Purpose                               |
-| ----------------- | ----------------- | ------------------------- | ------------------------------------- |
-| `build-artifacts` | Blacksmith 4 vCPU | `check-lint`, `code-size` | Builds dist and uploads artifact      |
-| `install-check`   | Blacksmith 4 vCPU | `check-lint`, `code-size` | Verifies `pnpm install` works cleanly |
+| Job               | Runner            | Depends on                    | Purpose                          |
+| ----------------- | ----------------- | ----------------------------- | -------------------------------- |
+| `build-artifacts` | Blacksmith 4 vCPU | `check-lint`, `code-analysis` | Builds dist and uploads artifact |
+| `release-check`   | Blacksmith 4 vCPU | `build-artifacts`             | Validates npm pack contents      |
 
 ### Tier 4+ — Tests and Platform
 
-| Job              | Runner             | Depends on                                   | Purpose                                                |
-| ---------------- | ------------------ | -------------------------------------------- | ------------------------------------------------------ |
-| `checks`         | Blacksmith 4 vCPU  | `check-lint`, `code-size`                    | TypeScript checks, tests (Node + Bun), protocol checks |
-| `checks-windows` | Blacksmith Windows | `build-artifacts`, `check-lint`, `code-size` | Windows-specific lint, tests, protocol checks          |
-| `macos`          | `macos-latest`     | `check-lint`, `code-size`                    | TS tests + Swift lint/build/test (PR only)             |
-| `android`        | Blacksmith 4 vCPU  | `check-lint`, `code-size`                    | Gradle test + build                                    |
+| Job              | Runner             | Depends on                                       | Purpose                                                |
+| ---------------- | ------------------ | ------------------------------------------------ | ------------------------------------------------------ |
+| `checks`         | Blacksmith 4 vCPU  | `check-lint`, `code-analysis`                    | TypeScript checks, tests (Node + Bun), protocol checks |
+| `checks-windows` | Blacksmith Windows | `build-artifacts`, `check-lint`, `code-analysis` | Windows-specific lint, tests, protocol checks          |
+| `macos`          | `macos-latest`     | `check-lint`, `code-analysis`                    | TS tests + Swift lint/build/test (PR only)             |
+| `android`        | Blacksmith 4 vCPU  | `check-lint`, `code-analysis`                    | Gradle test + build                                    |
 
-## Code-Size Gate
+## Code-Analysis Gate
 
-The `code-size` job runs `scripts/analyze_code_files.py` on PRs to catch:
+The `code-analysis` job runs `scripts/analyze_code_files.py` on PRs to catch:
 
 1. **Threshold crossings** — files that grew past 1000 lines in the PR
 2. **Already-large files growing** — files already over 1000 lines that got bigger
 3. **Duplicate function regressions** — new duplicate functions introduced by the PR
 
 When `--strict` is set, any violation fails the job and blocks all downstream
-work. On push to `main`, the code-size steps are skipped (the job passes as a
+work. On push to `main`, the code-analysis steps are skipped (the job passes as a
 no-op) so pushes still run the full test suite.
 
 ### Excluded Directories
@@ -109,26 +109,25 @@ The analysis skips: `node_modules`, `dist`, `vendor`, `.git`, `coverage`,
 **Bad PR (formatting violations):**
 
 - `check-format` fails at ~43 s
-- `check-lint`, `code-size`, and all downstream jobs never start
+- `check-lint`, `code-analysis`, and all downstream jobs never start
 - Total cost: ~1 runner-minute
 
 **Bad PR (lint or LOC violations, good format):**
 
-- `check-format` passes → `check-lint` and `code-size` run in parallel
+- `check-format` passes → `check-lint` and `code-analysis` run in parallel
 - One or both fail → all downstream jobs skipped
 - Total cost: ~3 runner-minutes
 
 **Good PR:**
 
 - Critical path: `check-format` (43 s) → `check-lint` (1m 46 s) → `build-artifacts` → `checks`
-- `code-size` runs in parallel with `check-lint`, adding no latency
+- `code-analysis` runs in parallel with `check-lint`, adding no latency
 
 ## Composite Action
 
 The `setup-node-env` composite action (`.github/actions/setup-node-env/`)
 handles the shared setup boilerplate:
 
-- Submodule init/update with retry (5 attempts, exponential backoff)
 - Node.js 22 setup
 - pnpm via corepack + store cache
 - Optional Bun install
@@ -141,7 +140,7 @@ This eliminates ~40 lines of duplicated YAML per job.
 
 ## Push vs PR Behavior
 
-| Trigger        | `code-size`                   | Downstream jobs       |
+| Trigger        | `code-analysis`               | Downstream jobs       |
 | -------------- | ----------------------------- | --------------------- |
 | Push to `main` | Steps skipped (job passes)    | Run normally          |
 | Pull request   | Full analysis with `--strict` | Blocked on violations |