chore(mcp): remove support for extension protocol v1

[yury-s]

All clients use protocol v2 by now. Remove the v1 handlers on both the relay and the extension side and collapse the abstractions that only existed to support two protocol versions:

• Relay: drop ExtensionProtocolV1 and the ExtensionProtocolHandler interface, always use ExtensionProtocolV2.
• Extension: fold the v2 protocol handler into RelayConnection, drop the eager (v1) pending connection path and the protocolVersion plumbing. The connect page now rejects clients requesting protocol version < 2.

[github-actions]

Test results for "MCP"

4 failed
❌ [chrome] › mcp/annotate.spec.ts:110 › should abort annotation when last screenshot is removed @mcp-macos-latest-chrome
❌ [chrome] › mcp/annotate.spec.ts:496 › should disengage annotate mode when --annotate client disconnects @mcp-macos-latest-chrome
❌ [chromium] › mcp/annotate.spec.ts:446 › should switch screencast to -s session on show --annotate @mcp-macos-latest-chromium
❌ [firefox] › mcp/http.spec.ts:104 › http transport browser lifecycle (isolated) @mcp-ubuntu-latest-firefox

7756 passed, 1249 skipped

---

Merge workflow run.

[github-actions]

Hi, I'm the Playwright bot and I took a first look at the failing CI.

🟢 The failures are all pre-existing flakes — this PR looks clear

The 4 failures are in mcp/annotate.spec.ts and mcp/http.spec.ts, none of which touch the extension-relay / protocol-v1 code this PR removes. Each one also flips verdict across many unrelated PRs in the CI results DB, so they're flakes, not regressions.

Details

This PR removes extension protocol v1 (relay cdpRelay*, packages/extension/*, tests/extension/*). The failures are in unrelated MCP browser flows and fail in a small fraction of runs across 123 distinct PRs each — the hard-rule proof of a flake.

Pre-existing flake / infra

• [chrome] › mcp/annotate.spec.ts:110 › should abort annotation when last screenshot is removed — flips across runs: 7/298 firefox, 4/305 chrome, 3/304 chromium, 2/307 webkit, 1/296 msedge failed, over 123 PRs.
• [chrome] › mcp/annotate.spec.ts:496 › should disengage annotate mode when --annotate client disconnects — 4/307 webkit, 3/304 chromium, 2/298 firefox, 1 chrome, 1 msedge failed, over 123 PRs.
• [chromium] › mcp/annotate.spec.ts:446 › should switch screencast to -s session on show --annotate — 5/298 firefox, 4/304 chromium, 3/305 chrome, 1/307 webkit failed, over 123 PRs.
• [firefox] › mcp/http.spec.ts:104 › http transport browser lifecycle (isolated) — 3/304 chromium, 2/307 webkit, 1 firefox, 1 msedge failed, over 123 PRs.

All four spread the failures across browsers/OSes and PRs the change can't be responsible for, and none exercise the removed v1 relay/protocol path. I'm a first pass, but the signal here is clean.

_{Triaged by the Playwright bot - agent run}