Add the authored High CPU alert-notebook template (#4223)

[erikdarlingdata]

Refs #4223.

Why

"High CPU" fires on both engines (SQL Server and PostgreSQL) and, like the other alert families, was falling back to a mechanical list of reads instead of getting an authored, forensic-shaped notebook.

What changes

AlertNotebookEndpoint.AuthoredTemplate("High CPU") now returns an authored/cpu template (version 1) with a fixed 7-cell layout:

#	Cell	Type	Detail
1	Header	header	incident/history summary
2	Status	status	current status text
3	SQL Server CPU	panel	cpu_utilization_stats / sqlserver_cpu_utilization, avg, line, hourly buckets
4	PostgreSQL CPU	panel	pg_cpu_utilization / pg_acu_utilization_pct, avg, line, hourly buckets
5	Top queries by CPU	read	get_top_queries_by_cpu (hours=24, top=10)
6	Top procedures by CPU	read	get_top_procedures_by_cpu (hours=24, top=10)
7	Scheduler pressure	read	get_cpu_scheduler_pressure

Both engines' CPU panels are always present — same "not applicable" convention the mechanical map already uses for this metric (the wrong-engine read/panel just comes back empty).

Both CPU measures are Gauge archetype (point-in-time percent readings), so their valid aggregates are avg/min/max, never count. The shared TimelinePanel core helper hardcodes aggregate: "count" for the per-event deadlock/blocking sources it was built for, so this template adds its own small CpuGaugeTimelinePanel helper (private to the new file) instead of widening the shared one — that would change what every other already-shipped family's panel emits. No annotation overlay is added: none of the available annotation sources (deadlocks, blocked-process reports, long-query completions, default-trace events, system_health events) is a CPU-specific marker, so the panels carry no annotations key.

get_top_queries_by_cpu, get_top_procedures_by_cpu and get_cpu_scheduler_pressure declare no limit param in the read catalog (the first two cap with top instead, the last has no row cap at all), so they're added to s_authoredLimitlessTrendReads alongside get_deadlock_trend.

Two pre-existing tests hardcoded "High CPU" as an example of a metric that stays on the mechanical fallback path; both are updated to use "Poison Wait" instead, which still routes mechanically:

• AlertNotebookAuthoredTemplateTests.AuthoredTemplate_NonAuthoredMetric_StaysMechanical
• AlertNotebookEndpointTests.AuthenticatedRequest_ReturnsTheDocumentedNotebookShape

AlertNotebookAuthoredTemplateTests.Budget_ReadCellsHaveLimitsExceptTheTrendRead_ComposedWindowsAreAtMost24h (part of the shared theories, data-driven off the registration table) is extended with the same top-instead-of-limit / no-cap exceptions for the three new reads.

Test plan

New file Darling.Tests/AlertNotebookTemplateCpuTests.cs (3 tests): routing to authored/cpu, the exact 7-cell list with both engines' panels titled and sourced correctly, and the no-matched-incident degrade (same cell count, still validates).

• RED confirmed against origin/dev (detached worktree): all 3 new tests fail — AuthoredTemplate("High CPU") returns null there (no registration row yet).
• GREEN on this branch, in-process on macOS (Darling.Tests.dll, no live PG needed — these are pure):
  • AlertNotebookTemplateCpuTests: 3/3 pass.
  • AlertNotebookAuthoredTemplateTests (shared theories, now covering "High CPU" via its MemberData): 40/40 pass.
  • AlertNotebookEndpointTests: 29/29 pass.
  • DocCommentHygieneTests: 77/77 pass.
• Both test projects build 0 errors / 0 warnings: Darling.Tests.csproj and Lite.Tests.csproj (-p:EnableWindowsTargeting=true). Both target net10.0-windows and cannot run here; CI decides them.
• Did not run the full repo test suite (time budget); the classes above are the ones this change touches.

CHANGELOG entry

SECTION: Added

ENTRY:

• High CPU alerts open an authored notebook: SQL Server and PostgreSQL CPU timelines, top queries/procedures by CPU, and scheduler pressure, instead of the mechanical read list. ([Add the authored High CPU alert-notebook template (#4223) #4420])

REF:
[#4420]: #4420