perf: gather all event listeners in parallel

[brendankenny]

this was some low hanging performance fruit discussed last week, getting all the event listeners for all nodes in the page in parallel rather than waiting on each command to finish. We aren't enabling/disabling domains or listening for events, so this should be safe.

Profile for the EventListeners gatherer doesn't look all that different but it does save e.g. 700ms (1300 -> 600) when running on www.chromestatus.com.

before:

after:

(purple block under 11750ms is all the commands now being sent at once. Remaining scatter of blocks is now just returning messages, not interleaved returning messages and sending messages)

[ebidel]

Small changes. Big/medium wins. 🏆

[brendankenny]

sneaky post-approval commit: @paulirish reminded me we had also discussed disabling websocket compression since our payloads are so tiny and the vast majority of the time in sendCommand and its callbacks is spent compressing and decompressing messages. So I've disabled it.

However, the win is basically imperceptible. The time spent per command has dropped, but the execution time of all commands was so small the reduction had little effect on overall execution time.

However however, the ws readme recommends:

The [permessage-deflate] extension is enabled by default but adds a significant overhead in terms of performance and memory consumption. We suggest to use WebSocket compression only if it is really needed.

and there may come a time where we need a gatherer that runs many commands in a row that can't be run in parallel, so leaving compression off as the new default makes sense.

[ebidel]

🚫 🏋️

[paulirish]

LGTM! so good