fix(app): fix browser layer notification error handling

[mjhuff]

Closes RQA-3108 (for the third time)

Overview

Fixes an issue in which browser layer error messages were dropped. The app shell sends MQTT error event messages on the ALL_TOPICS topic, but browser layer callbacks do not ever subscribe to the ALL_TOPICS topic, just their individual topic. This means when the app receives a disconnect event that is applicable to all topics, those topics do not default back to polling.

We know this fixes the ticket (all of the prior fixes are valid for this specific issue, too!), because you can see the error message at 2024-09-05T18:57:34.783Z in the Windows App Logs (attached in the ticket). The app, however, never defaults to polling, as evidence by the lack of network requests in the server logs after this time (only the ODD network requests show up).

Test Plan and Hands on Testing

CURRENT BEHAVIOR REPRO

• On the built version of the desktop app (or any other branch -- much easier to do this on the desktop than the ODD), run a protocol with a manual move, such as this one.
• Once at the manual move step, SSH into the robot and do a systemctl stop mosquitto.
• Verify in the app-shell that the Error - Error: read ECONNRESET message appears.
• Click continue on the manual move modal.
• You'll see the modal never dissapears.

PR FIX VALDIATION

• After switching to this branch, FULLY close out of the app and re-open(need to do this based on how notification connectivity works).
• Perform the above sequence of steps.
• After clicking continue, you should see the modal disappear!

Changelog

• Fixed the desktop/app not falling back to polling in some spots.

Risk assessment

• lowish. It's networking, which is always scary, but the fix is quite localized.

[sfoster1]

Makes sense to me!