feat(tracing): Allow empty array as tracingOrigins

[AbhiPrasad]

Fixes #5393

Allow for users to set an empty array as tracingOrigins option if they do not want to attach outgoing headers to any request. Previously we would just set the default tracing origins if users supplied an empty array.

This is technically a breaking change - but moves us toward the correct behaviour so I don't mind rolling this out in a minor. Thoughts?

[github-actions]

size-limit report 📦

Path	Size
@sentry/browser - ES5 CDN Bundle (gzipped + minified)	19.39 KB (0%)
@sentry/browser - ES5 CDN Bundle (minified)	59.99 KB (0%)
@sentry/browser - ES6 CDN Bundle (gzipped + minified)	17.96 KB (0%)
@sentry/browser - ES6 CDN Bundle (minified)	52.88 KB (0%)
@sentry/browser - Webpack (gzipped + minified)	19.73 KB (0%)
@sentry/browser - Webpack (minified)	64.21 KB (0%)
@sentry/react - Webpack (gzipped + minified)	19.75 KB (0%)
@sentry/nextjs Client - Webpack (gzipped + minified)	44.38 KB (-0.02% 🔽)
@sentry/browser + @sentry/tracing - ES5 CDN Bundle (gzipped + minified)	25.88 KB (-0.03% 🔽)
@sentry/browser + @sentry/tracing - ES6 CDN Bundle (gzipped + minified)	24.25 KB (-0.04% 🔽)

[Lms24]

Code-wise, this looks good to me ✅

This is technically a breaking change - but moves us toward the correct behaviour so I don't mind rolling this out in a minor. Thoughts?

So, I have a couple of thoughts on this:

• I agree that semantically, it makes a lot of sense to me that an empty array should simply not send trace data anywhere (+ span creation ofc). So from that PoV, the change is definitely going into the right direction.
• With the current behaviour, I think we're effectively going against the intuition of our users (albeit with a warning)
• Therefore, to me, this provides a good base of saying that this can be seen as a fix rather than "new" behaviour
• In terms of breakage, I made a quick GH code search and it turns out that not a lot of projects (other than sentry-js forks) actually use empty arrays. Ofc that's not the full picture but it at least somewhat tells us that the usage seems to be limited. Meaning, the impact shouldn't be too "bad".

We shouldn't forget though that,

a) this property will soon be deprecated in favour of tracePropagationTargets (where we should handle empty arrays just like this PR intends to)
b) This doesn't only influence trace data propagation but also span creation of outgoing requests (which we intend to separate from propagation with a new mechanism anyway). Not sure if this has a lot of implications for the severity of this change but I assume that it doesn't.

So IMO, let's go with this change and do it "right" in tracePropagationTargets from the start. In case we get a lot of complaints, we can still revert this and tell people to use some obscure string instead of an empty array to effectively stop sending trace data.

Let's see what the others think (we can also chat about it at daily)

[Lms24]

After discussing this at Daily, let's do this.

[AbhiPrasad]

Note - we'll have to shout this out in the changelog.