Replace the brittle tail -n 1 parsing with something that searches for the marker

[derekmisler]

Summary

Fix result marker parsing in the auto-issue-triage workflow. The previous tail -n 1 approach silently discarded valid agent results when LLM output contained trailing empty lines — which happened on the very first production run (actions/runs/22631502933), where the agent successfully diagnosed and fixed issue #1651 but the result was dropped.

Changes

• Replace tail -n 1 | tr -d '[:space:]' with grep -oE 'RESULT:(NEEDS_INFO|FIXED|NO_CHANGES)' | tail -n 1 to search the entire output for the result marker instead of relying on it being the literal last line
• Update the warning message to reflect the new search-based approach

Test plan

• Re-run the triage workflow on an issue with kind/bug label — verify the result marker is correctly parsed even with trailing empty lines in agent output
• Verify RESULT:NEEDS_INFO, RESULT:FIXED, and RESULT:NO_CHANGES paths still work correctly
• Verify that when no marker is present, the fallback action=none is still set