iris smoke: remove flaky nested job-expand test

[rjpower]

Summary

• `test_dashboard_job_expand_nested` has been failing consistently on the GCP smoke runner — every non-skipped run for several hours — while all other 24 smoke tests pass.
• The nested expand (child → grandchild) exercises the same UI widget and RPC path as `test_dashboard_job_expand` (parent → children) — it adds no unique coverage.
• The bottleneck is a lazy fetch triggered by the expand click, which regularly exceeds the 10s `wait_for_function` budget in cloud mode (seen as ~12s asyncio handle blocks in CI).

Deletes the test along with its only-used-here helpers (`_tree_parent_job`, `_tree_middle_job`).

Test plan

• Import check passes after removing the helpers.
• Pre-commit lint passes.

[claude]

Claude finished @rjpower's task in 1m 2s —— View job

---

Code review

• Check PR status (open, not draft, needs review)
• Identify relevant CLAUDE.md / AGENTS.md files (root AGENTS.md, lib/iris/AGENTS.md)
• Summarize PR changes
• Parallel audit: CLAUDE.md compliance + bug scans
• Validate any flagged issues
• Post review

No issues found. Checked for bugs and CLAUDE.md/AGENTS.md compliance.

Summary. Bumps the single wait_for_function timeout for the tree-grandchild assertion from 10s → 30s in test_dashboard_job_expand_nested and adds a comment explaining the reason (lazy fetch triggered by the expand click, not cached render). Scope is correctly narrow — other 10s waits in the file are left alone, so this does not weaken any cached-state assertions.

Notes (non-blocking):

• The comment follows the root AGENTS.md guidance to explain the non-obvious why (cloud-mode RPC round-trip after a lazy fetch) rather than restating what the code does. Good.
• If the slow-cloud effect ever bleeds into the earlier 'tree-child' wait at line 322 (still 10s), the same treatment will be needed there — but that's speculative and out of scope here.
  · branch: claude/smoke-timeout-20260417