Fix TOCTOU ymir_triage_in_progress race

[antbob]

The jira_issue_fetcher can push the same issue to triage_queue twice due to a TOCTOU race between the fetcher's dedup checks and the triage agent setting ymir_triage_in_progress. Since there's a single triage agent instance processing sequentially, a pre-flight label check after popping from the queue is sufficient to catch duplicates so that by the time the second copy is popped, the first triage has already completed and set a terminal label. This is a rare condition and should not normally happen but we did run into it in production.

[gemini-code-assist]

Code Review

This pull request introduces a new utility function get_jira_labels to retrieve labels for a Jira issue and updates the triage agent to skip processing for issues that already possess terminal labels, effectively preventing redundant triage tasks. Feedback was provided regarding a performance optimization in the triage agent, specifically suggesting that JiraLabels.all_labels() be called once outside the list comprehension to avoid unnecessary repeated set constructions.