ref(node): Streamline dataloader instrumentation

[mydea]

This streamlines the vendored dataloader instrumentation:

• Use Sentry.startSpan instead of tracer APIs
  • this allows us to streamline a bunch of things, e.g. using the built-in onlyIfParent option
• Remove instrumentation options we hardcode anyhow (we always use onlyIfParent)
• Streamline types to what we actually need
• Fix linting errors (remove eslint-disable comment)
• Add tests covering all methods, actually
• Moved custom span attributes etc. we had to hackily set in on('spanStart') directly into the instrumentation

This is overall much simpler, as we use the built-in error handling etc. from Sentry.startSpan.

Closes #20725

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 070c3c0. Configure here.}

[cursor]

Chained expects assume envelope order

Medium Severity

The test chains four .expect({ transaction: … }) callbacks, but the integration runner matches incoming envelopes with expectedEnvelopes.shift(), so the first callback always runs on the first transaction received—not necessarily GET /load. Four back-to-back HTTP requests can emit transactions in a different order than the expects, causing flaky failures when the wrong callback validates the wrong route.

 

^{Triggered by project rule: PR Review Guidelines for Cursor Bot}

^{Reviewed by Cursor Bugbot for commit 070c3c0. Configure here.}

[mydea]

this should not be possible to flush in a different order