Properly handle the Azure Recognizer shutdown after audio ended.

[pragmatrix]

After we stopped audio, the following appeared in the log:

2026-08-05T12:09:13.048499Z ERROR context_switch::context_switch: Graceful shutdown period expired after waiting for 3000ms

Even though this situation is handled by sending an empty audio frame in azure-speech-sdk, the session is restarted and the event stream stays open.

The updates in the azure-speech-sdk submodule won't restart the session and wait for it to end and then ends the event stream.

[Copilot]

Pull request overview

Adds an (ignored) Azure integration regression test to ensure Azure transcription sessions stop cleanly after audio input ends, and augments Azure service logging to help diagnose recognizer lifecycle behavior. Also updates workspace metadata and dev-dependency features needed for the new test’s tracing configuration.

Changes:

• Added an ignored Tokio integration test that starts Azure transcribe, streams short audio, then asserts a timely Stopped event after Stop.
• Added additional Azure transcribe service logging around connection and per-event reception.
• Enabled tracing-subscriber’s env-filter feature for tests and bumped workspace version.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
src/tests.rs	Adds ignored Azure shutdown regression test and Azure params helper.
services/azure/src/transcribe.rs	Adds connection debug logs and per-event trace logging for Azure recognizer.
Cargo.toml	Bumps workspace version and enables tracing-subscriber env-filter feature in dev-deps.