Surface request-stream defects in Zio client streaming

[ghostdogpr]

Problem

In ZioClientBackend.clientStreamingRun, the request-sending effect is guarded with catchAll:

val send: UIO[Unit] = sendAll.catchAll { e =>
  ZIO.succeed(call.cancel("Error sending requests", e)) <* listener.promise.fail(e)
}

catchAll recovers only the typed error channel (E = StatusException). If the user's request stream raises a defect (Cause.Die) rather than a typed failure — a map/mapZIO that throws, ZStream.fromIterator over a throwing iterator, etc. — the handler never runs. So:

• call.cancel(...) and call.halfClose() never execute → the server keeps waiting for messages and never sends onClose;
• listener.promise.fail(e) never runs → promise.await is never completed;
• the sender is forkDaemon'd, so its death is not propagated to the parent.

With CallOptions.DEFAULT (no deadline) the call hangs indefinitely, leaking the ClientCall and the awaiting fiber. The fs2 (handleErrorWith) and Kyo (Result.Panic) backends already handle this case; Zio was the outlier.

Fix

Use catchAllCause, which recovers both typed failures and defects. cause.squash yields a Throwable for call.cancel, and promise.failCause(cause) preserves the original cause (typed or defect). The normal typed-error path is unchanged.

Test

Added a regression test that runs a client-streaming call whose request stream dies with a defect. Pre-fix it hangs forever; post-fix it completes in ~200ms. The test uses @@ TestAspect.withLiveClock so the .timeout guard runs against real time rather than the frozen TestClock.