[actions] possible regression, jobs are not complete but canceled instead

I started to see a change in the testflinger action behavior recently.

For example,

```
+ exit 1
2025-05-09 11:31:04,608 hinopio WARNING: DEVICE CONNECTOR: Tests failed, rc=1
2025-05-09 11:31:04,609 hinopio INFO: DEVICE CONNECTOR: END testrun

*************************************************

* Starting testflinger cleanup phase on hinopio *

*************************************************
Cleaning up container if it exists...
hinopio
2025-05-09 11:31:11,557 hinopio INFO: DEVICE CONNECTOR: MAAS: Reading node's block device information
2025-05-09 11:31:13,521 hinopio INFO: DEVICE CONNECTOR: MAAS: Releasing node hinopio
```

^^^ a test has failed at this point and MAAS machine is going to be released.

```
}2025-05-09 11:31:49,541 hinopio INFO: DEVICE CONNECTOR: MAAS: Successfully released node hinopio
Error: The operation was canceled.
Run echo "::group::Cancel job"
echo "::group::Cancel job"
testflinger --server $SERVER cancel $JOB_ID
echo "::endgroup::"
shell: /usr/bin/bash --noprofile --norc -e -o pipefail {0}
env:
SERVER: https://testflinger.canonical.com/
JOB_ID: d5d5133f-d4cf-4993-b1cf-4ed78b013f91
Cancel job
```

^^^ successfully released machine, then `Error: The operation was canceled.`