E2E: fix the concurrent editor routing spec's attach gate and connection deadlock

[duolahypercho]

Root cause

workflow.spec.ts "routes concurrent editor calls to the owning page and session" timed out intermittently under the full suite (7/12 in the PR45 authoring run, reproduced locally on merge trees and on harness's direct repeat-3). Two compounding problems:

1. Attach registration gate. The spec polled editor_attach on request (outgoing), then immediately fired editor_tool_call. Core registers the session's editor owner while handling the attach, so the tool call could arrive before an owner existed and hang until the 60s test timeout instead of failing with a reason. requestfinished fires after the response is back, which is the point where the attach is actually applied.
2. Browser connection pool deadlock. The two long-lived editor_tool_call invocations were issued through each page's own fetch from page.evaluate. Under the 9-worker suite these page-owned connections can deadlock the browser's pooled connection set, which is also why the timeout surfaced as page.evaluate: Test ended / browserContext.close in cleanup. Routing only those two RPCs through Playwright's typed APIRequestContext removes the page-origin connection dependency; the routing assertions are unchanged.

Changes

• client/e2e/workflow.spec.ts: capture editor_attach on requestfinished instead of request; add callRpcFromRunner and use it for the two concurrent editor_tool_call invocations only. No product/App/SSE/permission changes.

Verification

• npx tsc -b --noEmit: pass
• pnpm test: 904 passed
• pnpm build: pass
• Targeted pnpm test:e2e -- --grep "routes concurrent" --repeat-each=3: 3/3 pass
• Full pnpm test:e2e (71 tests): 71 passed