Memoize Playwright page target IDs

[rgarcia]

Summary

• cache successful Playwright page-to-CDP target ID lookups in a WeakMap
• coalesce concurrent discovery and prevent pre-reset work from repopulating the cache
• refresh cached IDs on resolver retries and reset them on browser disconnect
• build a per-attempt pageByTargetId index for the tab-to-page join
• serve each resolution attempt with a single browser-level CDP session (the active-tab listing and every tab-to-page join share one session instead of opening one per active tab)
• add focused Node tests for reuse, refresh, failures, concurrency, and reset behavior

Stack

Based on #347. Replaces #352/#355 after the resolve_active_page request option was dropped from the stack.

Validation

• make test-runtime (9/9)
• go vet ./...
• Playwright daemon esbuild bundle build

[masnwilliams]

reviewed the stacked layer at 16d6b3a. one root CDP session per attempt removes the nested session churn, while the weak cache is refreshed on retry and reset on browser disconnect/timeout. the cache tests cover reuse, failed discovery, refresh, in-flight work, and reset.

validation beyond BugBot: runtime tests and the full non-e2e race suite pass; both Dockerfiles pass build checks; the bundled daemon passed the deterministic cross-context case, 30 rapid cross-origin transitions with no failures/wrong pages, page recreation, and an unmatched raw-CDP tab. with 41 pages, warm active-first/active-last resolution was ~10ms in both positions.

this PR targets the lower stack branch, so the main-targeted image/server workflows have not run on this layer yet. please let those run after #347 merges and this retargets to main, before merging/releasing. approved subject to that normal CI gate.