Implement end2end tests

[camillobruni]

• Copy over package.json from Speedometer3
• Implement npm test:chrome, npm test:firefox, npm test:safari
• Add better support for custom worstCaseCount and iterationCount to lower test duration
• Pass along iterations to Benchmark.validate to fix unipoker expectations
• • Add JetStreamReady and JetStreamDone events to the Driver for easier testing

[kmiller68]

This seems like it will always set startDelay to undefined when the there's no startDelay URL parameter. I think that's fine for testIterationCount and testWorstCaseCount since they'd always be undefined on the web as it is. Although maybe it's worth changing there too so folks aren't surprised?

[camillobruni]

I hoped to make the initialization explicit by the lines above globalThis.startDelay ??= undefined; at least for me that's a bit more consistent across all configuration params.

Medium term I'd like to adapt the Params object from Speedometer (this also allows us to easily copy over the ?developerMode UI).

Maybe I missed your point here, but when running via the shell we now also always initialize startDelay=undefined.
With the html runner we'd initialize to undefined by default (and set it again to undefined if there is no matching url param)

But generally not feeling strongly about this one :)

[kmiller68]

I think the issue is when you do index.html?report=true startDelay ends up as undefined rather than 4000. If you moved the report check below this and changed it to if (shouldReport) globalThis.startDelay ??= 4000 I think that would fix it.

Safari, at least, relies on the startDelay to make sure the browser has calmed down after launching enough to prevent noise.

[camillobruni]

ok, then I understood correctly.
Keeping the old behavior with report=true sg.

[camillobruni]

I've now merged in the latest PR from speedomter to use the local-web-server package so we don't have to write our own little webserver.