fix: repair pre-existing watcher_pane tests

[ProtocolWarden]

Summary

Repairs 9 pre-existing failures in tests/test_watcher_pane.py, deciding per-failure whether the test was stale or the code regressed.

Stale tests (code was correct, tests updated)

• Banner copy was intentionally title-cased / reworded: "All Systems Nominal", "SwitchBoard Offline", "Global Gate at Cap", "Queue Depth", "Stabilizing". Updated 5 assertions.
• test_exec_budget_reads_usage never isolated _resource_gate(), which reads the real on-disk OC config; gate caps shadow the env caps the test sets. Added a monkeypatch stubbing the gate so the env-override path is exercised.
• test_critical_sorts_before_warning_sorts_before_info had a premise that conflicts with the intentional "just started" INFO pinning (documented in code). Split into a pure severity-order test (started_at=0) and a test documenting the pinning.

Real code regression (code fixed)

• _allocate_section_rows did no overflow scaling and over-allocated far past available_rows (42/31 vs a cap of 10). Added proportional down-scaling for expanded sections on overflow with a min(3, natural) per-section floor. Natural-fit, size_mult, collapsed, and empty paths preserved.

Test plan

• tests/test_watcher_pane.py: 27 passed
• Full suite: 132 passed, 3 failed (pre-existing cxrp schema_version 0.2-vs-0.3 mismatch in tests/test_cxrp_capture.py, unrelated)
• Custodian: 0 findings | clean (pre-push hook passed)

Files touched: src/operator_console/watcher_status_pane.py, tests/test_watcher_pane.py, .console/log.md only. launcher.py / bootstrap.py untouched.

🤖 Generated with Claude Code