fix(app-review): remediate dashboard tab review findings across all f…

[AdityaMandiwal]

…our tabs

Address all findings from the code review of the Job Clusters, All-Purpose, Instance Pools, and Pipeline Compute tabs plus the shared backend routers (see docs/plan_app_review_fixes.md, Phases 0-8).

Backend (Phase 1):

• Inverted date ranges now return 400 (not 500) on the four affected endpoints
• Genericize client-facing error responses; log full tracebacks server-side
• Remove the unauthenticated debug/test endpoints

Frontend correctness (Phases 2-3):

• Sanitize LLM output via react-markdown (AnalysisMarkdown) - removes XSS surface
• Fix React key collisions (job fragment, pipeline composite key)
• Add error states to all SummaryCards; key-remount tables on filter change to fix pagination-reset race and stale expansion state
• Fix date off-by-one in negative-UTC zones; dialog onOpenChange guard; a11y

Per-tab (Phases 4-7):

• Job Clusters: honest server-side sort, plain-text job-level Other cell, date-driven coverage trend, shared HIGH_COST_USD + runs "show all"
• All-Purpose: debounce-safe sub-tab switch, single Tabs root, popstate sync, By-User Other affordance + corrected help text
• Instance Pools: dynamic cloud label, per-pool day expansion, memoized sort
• Pipeline: gate analysis on resolved details, reset workloads on range change, OR chip copy, NaN-safe currency

Polish (Phase 8):

• Remove unused imports, stable table keys, per-section summary skeletons, client-side start>end date guard