fix(ruvector): bundle ONNX runtime into dist/ on build (#354)

[ruvnet]

Summary

• Closes ONNX WASM files not bundled on clean insteall #354 — the published tarball shipped only a 1-line dist/core/onnx/pkg/package.json and none of the ONNX WASM runtime files. Every clean install failed at initOnnxEmbedder() with Error: ONNX WASM files not bundled. The onnx/ directory is missing.
• Same root cause as CLI commands that take a <database> path are broken on freshly-created DBs (and embed text fails on missing ONNX bundle) #417 (part 2). CLI commands that take a <database> path are broken on freshly-created DBs (and embed text fails on missing ONNX bundle) #417 still has CLI-handler bugs to fix on top of this; that work continues in a separate PR.

Root cause

"build": "tsc && cp src/core/onnx/pkg/package.json dist/core/onnx/pkg/"

tsc only emits compiled .ts output (no allowJs), so the JS sources under src/core/onnx/ (the wasm-bindgen .wasm payload, _bg.js, type defs, LICENSE, sibling loader.js) never reach dist/. The script copied a single package.json and called it a day.

Fix

• Replace the single-file copy with scripts/copy-onnx-assets.js, a Node-portable recursive copy (no cp -r, works on Windows).
• Skip dotfiles (e.g. transient .claude-flow/ agent metadata) and node_modules/ so they don't leak into the published artifact.
• Sanity-check that the canonical runtime files (*_bg.{js,wasm}, *.js, loader.js) landed where onnx-embedder.js looks for them; fail the build loudly if not.

Proof

On a fresh build of the current 0.2.25 working tree (Node 22.22.2):

$ rm -rf dist/core/onnx && npm run build
> tsc && node scripts/copy-onnx-assets.js
copy-onnx-assets: 10 ONNX runtime file(s) staged under dist/.

$ ls dist/core/onnx/pkg/
LICENSE                                  ruvector_onnx_embeddings_wasm_bg.js
loader.js                                ruvector_onnx_embeddings_wasm_bg.wasm
package.json                             ruvector_onnx_embeddings_wasm_bg.wasm.d.ts
ruvector_onnx_embeddings_wasm.d.ts       ruvector_onnx_embeddings_wasm.js

$ npm pack && tar -tzf ruvector-0.2.25.tgz | grep -c 'dist/core/onnx/pkg'
8

# Clean install in /tmp:
$ node -e "const {isOnnxAvailable} = require('ruvector/dist/core/onnx-embedder'); console.log(isOnnxAvailable())"
true

Tarball grew from 825 KB → 3.1 MB; that's the 7.4 MB MiniLM WASM payload (uncompressed) compressing to ~2.3 MB on top of the 825 KB baseline — expected since this is the file the consumer was previously missing.

Test plan

• npm run build lands all 10 ONNX runtime files under dist/core/onnx/.
• npm pack includes them in the tarball (grep -c onnx/pkg = 8).
• Clean install of the packed tarball makes isOnnxAvailable() return true (was false/throw before).
• Build does not leak transient .claude-flow/ dotfiles into the artifact.
• Script doesn't depend on cp (portable to Windows).
• Optional follow-up: extend verify-dist.js to assert the ONNX runtime files exist (best done after fix(ruvector): verify-dist guards package.json entrypoints (#376) #433 lands to avoid merge conflicts).

Scope notes

• This PR only fixes the bundling (ONNX WASM files not bundled on clean insteall #354). It does not touch the CLI handler bugs reported in CLI commands that take a <database> path are broken on freshly-created DBs (and embed text fails on missing ONNX bundle) #417 — those are a separate fix.
• Once this lands, the loader change in fix(ruvector): Node 22 LTS-compatible ONNX WASM loader (#323) #432 (Node 22 LTS .wasm extension fix, ONNX embedder fails on Node 22 LTS with 'Unknown file extension .wasm' #323) can be exercised end-to-end on a clean install.

🤖 Generated with claude-flow