fix(sentry): Bind hub to lazy response streams

[jan-auer]

When an axum handler returns a streaming body via Body::from_stream(), the IntoResponse conversion runs inside the handler (within the sentry middleware), but the body's frames are polled by hyper after middleware has unwound. This means Hub::current() inside a stream producer resolves to whatever hub the tokio worker thread happens to have — not the request's hub.

Three endpoints are affected: object_get, complete (multipart), and batch.

We introduce SentryStream<S>, a stream wrapper similar to SentryFuture in the Sentry SDK. It binds a provided hub to the stream.

A blanket middleware approach was considered, but did not work: axum's Body is type-erased, so there is no way to know whether it is a stream. Wrapping unconditionally through the middleware would strip the size_hint from all buffered responses, causing hyper to fall back to chunked transfer encoding.

This utility is a candidate for upstreaming into the sentry Rust SDK in a follow-up, alongside SentryFuture and SentryFutureExt.

Alternative to #526.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.35%. Comparing base (824c05a) to head (acd434c).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #524   +/-   ##
=======================================
  Coverage   87.35%   87.35%           
=======================================
  Files          87       88    +1     
  Lines       14086    14103   +17     
=======================================
+ Hits        12305    12320   +15     
- Misses       1781     1783    +2     

Components	Coverage Δ
Rust Backend	92.16% <100.00%> (+0.01%)	⬆️
Rust Client	79.89% <ø> (ø)
Python Client	88.70% <ø> (ø)

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jan-auer]

This is one of the two locations where we now bind the hub.

To expand on the description for why this has to be done here: We could technically import http_body and then pipe the size hint and other methods through from the underlying stream. Still, it means that we now add another box around every response, and we lose any form of optimization that axum might have on eager responses.

As long as we only have two places where we stream responses back, I think this is fine. Once we have more places, we'll have to reassess this.

[jan-auer]

Moved back to draft as I prefer the alternative PR.