Add AI Patronus browser memory example

[Sparkah]

What changed

• Adds examples/ai-patronus-browser-memory, a Manifest V3 Chrome extension plus local FastAPI memory bridge.
• The extension can save the current page on explicit user request and recall saved pages semantically.
• The bridge calls SIE through the OpenAI-compatible /v1/embeddings endpoint and stores vectors locally in JSON.
• Adds the example to examples/README.md.

Why

This shows how SIE can power browser-agent memory without automatically uploading browsing history or requiring a hosted vector database. The default path is local SIE Docker plus local storage; optional provider keys are kept out of source and configured by the user.

Validation

• node --check on extension JavaScript files
• python3 -m py_compile on memory bridge scripts
• python3 -m json.tool on manifest.json and sample data
• secret-pattern scan over the packaged example
• private endpoint scan over the packaged example
• git diff --cached --check before commit

Notes

This is a draft PR so the Superlinked team can advise on preferred example scope, asset expectations, and whether they want the optional non-SIE provider hooks kept or trimmed.

[Sparkah]

Smoke test update from local branch:

• Extension JavaScript syntax checks passed with node --check.
• Memory bridge Python files compile with python3 -m py_compile.
• manifest.json and sample_pages.json validate with python3 -m json.tool.
• Secret-pattern and old private-endpoint scans found no matches in the packaged example.
• End-to-end memory bridge smoke passed using a local SIE-compatible /v1/embeddings stub:
  • POST /ingest stored two page memories with vectors.
  • POST /query returned scored recall results.
  • local JSON memory store contained the expected embedded records.

I did not run a full Docker SIE smoke because Docker CLI is installed but the local Docker daemon is not running in this session. The stub smoke validates the extension-facing bridge API and the OpenAI-compatible embeddings response shape.

[Sparkah]

Hosted SIE smoke update (using tester access supplied separately; no endpoint or credential is committed):

• Authenticated GET /readyz, GET /v1/models, and GET /openapi.json returned HTTP 200.
• The first POST /v1/embeddings calls returned the expected 503 provisioning response with Retry-After: 60; after the pool warmed, the same request succeeded.
• sentence-transformers/all-MiniLM-L6-v2 returned two 384-dimensional vectors.
• The full Patronus bridge flow passed against the hosted engine using synthetic data:
  • two POST /ingest calls stored local page memories with embeddings;
  • POST /query returned two scored results;
  • the privacy-related page ranked first for the privacy/browser-memory query;
  • the local JSON store contained both expected 384-dimensional vectors.

No real browser history or page data was used. The existing SIE_URL / SIE_API_KEY environment configuration works as written, so this smoke did not require a code change.