perf(registry): use weak ETag in express adapter

[ricardo-devis-agullo]

Change: default the express adapter to weak ETag (the Express default).

Why: packages/oc/src/registry/domain/http-server/express-adapter.ts explicitly set app.set('etag', 'strong'). No consumer relies on strong semantics (verified below). Weak keeps conditional-request/304 support with the W/ prefix and weak comparison, and restores the Express default. No static-file handling touched (registry has no express.static usage; static assets are served elsewhere).

No CPU saving is claimed: the etag package SHA-1 hashes string/Buffer bodies in both weak and strong modes (verified against its source -- weak only adds the W/ prefix and relaxes comparison). Fully skipping ETag generation for JSON responses would be a separate behavior change and is out of scope for this PR.

Consumer check (no reliance on strong ETags):

• grep If-None-Match/if-none-match across packages/: only oc-fastify-server-adapter tests send it.
• grep ETag/etag in packages/oc/src: only the adapter setting, plus unrelated refreshInterval hits and a 304 passthrough in registry/routes/helpers/get-component-fallback.ts (forwards client headers to the fallback registry; clients don't send validators).
• oc-client-browser sends no If-None-Match; batch POST (registry/routes/components.ts) and render routes return res.json with no conditional handling.

Fastify parity (not changed): packages/oc-fastify-server-adapter/src/index.ts:240 registers @fastify/etag with { weak: false, replyWith304: true } (strong), asserted in test/index.test.ts:43-72 and test/registry-node.ts:242-249. Left untouched to keep this PR minimal; a parity follow-up can flip weak:true plus those assertions if desired.

Verify headers:

• dist adapter check: app.get('etag') === 'weak'; GET /component -> 200 ETag: W/"b-..." (W/ present); raw-http conditional GET with If-None-Match -> 304.
• lock-in unit test in packages/oc/test/unit/registry-domain-http-server.js asserts adapter.native().get('etag') === 'weak'.
• unit: npx mocha test/unit/registry-domain-http-server.js (packages/oc) -> 3 passing.
• build: npm run build (turbo) -> 11 successful; biome check on touched files clean.

Bench: not run. bench:quick is 3 scenarios x 3 repetitions x 15s per branch (5+ minutes for both branches), and any RPS/p95 delta would be noise given both etag modes hash identically -- reporting numbers would be misleading.