Fix the alert notebook's collector-freshness check (#4378)

[erikdarlingdata]

Refs #4378.

Why

The alert-notebook endpoint's freshness check decides between "No resolution recorded" and "Unknown (not collected since …)" for a metric whose alert history gave no direct answer. It had three problems: it counted ANY collector run including ERROR rows as evidence the instrument was up, it ran without the sibling reads' command timeout, and its failure path always logged 0 ms elapsed regardless of how long the read actually took.

What changes

• The freshness query now filters to status = 'SUCCESS' (see Freshness evidence below).
• The freshness command sets CommandTimeout = McpCommandDeadlines.ReadSeconds (20s), the same budget the sibling MCP reads already use.
• Both the freshness read and the status-history read now measure elapsed time with a Stopwatch and report ElapsedMilliseconds on failure instead of a hardcoded 0.
• Reworded a stray internal comment marker in AlertNotebookAuthoredTemplateTests.cs to cite #4223 only.
• Test-only, alongside this fix: AlertNotebookAuthoredTemplateTests' shared theories (validate / known-dispatch / compose-catalog / budget / binding / viz) were driven off a hand-written two-metric list (BothTemplates()). That list is now AllAuthoredMetrics(), generated by flattening AlertNotebookEndpoint.s_authoredTemplates, so a future authored family (Alert notebooks: authored templates for the remaining families, including custom rules and analysis findings #4223) is covered by these shared theories the moment its registration row lands, with no edit to this test file. The routing theory is now similarly data-driven, asserting each registered metric routes to its own row's Entry.Id.

Recovery edges

The AG family and the Server Unreachable/Restored pair fire under their own metric names rather than through a resolution-title alias, so StatusFromHistory's arm 1 (which only consulted DarlingTriageEndpoint.ResolutionAliases) never reached them — a notebook opened against "AG Replica Disconnected" could not read "Resolved at T" even when "AG Replica Reconnected" arrived later.

Added a small notebook-only pair list (AlertNotebookEndpoint.NotebookRecoveryEdges) consulted by arm 1 alongside ResolutionAliases:

• AG Replica Disconnected → AG Replica Reconnected
• AG Database Suspended → AG Data Movement Resumed
• AG Sync Fell Behind → AG Sync Recovered
• Server Unreachable → Server Restored

This is deliberately NOT added to ResolutionAliases itself, because that list also builds DarlingTriageEndpoint.SectionsByMetric, and every metric above already has its own entry there — adding them to ResolutionAliases would define the same section-map key twice.

AG Failover is left out on purpose: it is an event (a role change), not a condition that clears, so it has no natural recovery edge.

Freshness evidence

Only SUCCESS counts as evidence the collector reached the target. SKIPPED was dropped from the freshness predicate: it can be written before any query reaches the target — Lite/Services/RemoteCollectorService.cs's user-cancelled-MFA catch sets status = "SKIPPED" at the point the sign-in is cancelled, never having contacted the server. Darling's own per-run classifier, EnumeratedCollectorDriver.ClassifyReturnedRun, never returns SKIPPED either, so the status carries no server-reachability information on either code path.

Test plan

• AlertNotebookEndpointTests, AlertNotebookAuthoredTemplateTests, DocCommentHygieneTests — all pass in-process on this machine (32, 77, 29 total respectively, 0 failed).
• New pure pins in AlertNotebookEndpointTests.cs: CollectorFreshnessSql contains the status predicate and excludes 'ERROR'; a source-text pin confirms CommandTimeout = McpCommandDeadlines.ReadSeconds is set and that no Report(..., 0, ex) call remains for either read; three recovery-edge pins (AG reconnect, Server Restored, AG Failover absent from the pair list).
• New live class AlertNotebookCollectorFreshnessLiveTests (3 facts), against my own throwaway container (timescale/timescaledb:2.30.1-pg18, removed after this run): only-ERROR-after-anchor reads Unknown; SUCCESS-after-anchor reads "No resolution recorded"; only-rows-before-anchor reads Unknown. All 3 GREEN on this branch.
• RED confirmed on origin/dev (084cfb3) in a detached worktree: both the pure and the live test files fail to COMPILE against the old code (ResolveStatusAsync, CollectorFreshnessSql, and NotebookRecoveryEdges don't exist there — ResolveStatusAsync was private and is now internal for the live test to call it directly).
• Both Darling.Tests.csproj and Lite.Tests.csproj build 0 errors with -p:EnableWindowsTargeting=true.
• LivePostgresCollectionHygieneTests.EveryClassUsingTheSharedStore_IsSerializedOrDocumentsWhyNot fails on this Mac with a PresentationFramework load error — pre-existing platform gap unrelated to this change (it reflects over all loaded assemblies, which includes a WPF-referencing assembly this environment cannot load); left for CI, which runs on Windows.
• Darling.Tests/Lite.Tests target net10.0-windows and cannot fully run here; CI decides the rest of the suite.

CHANGELOG entry

SECTION: Fixed
ENTRY:

• Alert notebook: the collector-freshness check now counts only successful collection runs ([Fix the alert notebook's collector-freshness check (#4378) #4416]) - it previously treated a collector's own failed runs as proof it was still fresh, used no read timeout, and always logged a zero elapsed time on failure. It also now recognizes AG and Server Unreachable/Restored recovery pairs when resolving a notebook's status.
  REF:
  [Fix the alert notebook's collector-freshness check (#4378) #4416]: Fix the alert notebook's collector-freshness check (#4378) #4416