fix(publish-metrics): otel tracing bug

[InesNi]

Description

Fix for OTel traces arriving inconsistently (in number and format) to observability platforms when running in Fargate

Context

Two different issues were observed, and usually at least one would happen:

1. All traces properly formed with the correct attributes and status, but some traces missing
2. Traces not properly formed - missing parent root spans, incorrect span status (errors not set on spans)

Note: the issues would not happen on local runs, and issue 2. seems to only happen on Datadog.

The issues seem to be a result of multiple factors:

• The wait for pending traces was not implemented in Playwright tracing possibly leading to triggering the traceProvider shut down prematurely.
• In large tests with many traces the maxExportBatchSize and eventually the maxQueueSize could be reached in which case spans would be dropped

Solution

This PR addresses the mentioned factors that seem to be causing the issues:

1. Waiting for pending traces is implemented in the Playwright tracing itself
2. The defaults for maxExportBatchSize, maxQueueSize and the scheduledDelayMillis were increased, and the settings exposed to user for configuration if needed (this would allow users to adjust the number of traces that a batch can take before dropping them)
3. Additional 5sec wait is added prior to shutting down the traceProvider in order to wait for the exporters flush.

[hassy]

Is there a way to reduce the flush period of the Datadog agent?