Investigate settings to reduce 499 Client Closed errors from gitlab-runner

While investigating gitlab-com/gl-infra/production#20469, I've noticed one of the common reasons why CI jobs appear to be running but not actually running is that Cloudflare sent a 499 Client Closed response to the runners.

Over the last 14 days, we see a steady stream of 499 Client Closed errors just to the /api/v4/jobs/request endpoint. It's also curious this is happening mostly on HTTP/1.1:

This was originally brought up years ago in gitlab-com/gl-infra/production-engineering#10606 (closed) with no resolution.

In a few examples, I noticed a one-second timer between the EdgeStartTimestamp and EdgeEndTimestamp:

"EdgeStartTimestamp": "2025-11-13 01:02:23.000000 UTC",
"EdgeEndTimestamp": "2025-11-13 01:02:24.000000 UTC",

I also noticed that the Runner sets ExpectContinueTimeout to 1 second: https://gitlab.com/gitlab-org/gitlab-runner/-/blob/e6451fb2fd16c97e4a0d8775ce559af7723d7b6a/network/client.go#L204

I think we should set this value 0.

There may be other settings we need to tune.