Skeleton Loaders, Helm Auto-Release & Performance Fixes

Added

• Skeleton loaders and loading overlays: all dashboard pages now show content-shaped loading states instead of blank spinners
  • New Skeleton, SkeletonTable, and TableLoadingOverlay components (src/lib/components/ui/skeleton/)
  • Directional shimmer animation via @keyframes shimmer using design tokens — works in light and dark mode, disabled for prefers-reduced-motion
  • Initial load (no data yet): animated skeleton rows mirroring the page layout — stat cards on /dashboard, project cards on /dashboard/projects, table rows on search, traces, errors, admin tables, incidents, alerts history, and members
  • Re-fetch (filter change, pagination): existing content dims with a translucent overlay and centered spinner, preventing layout shift and context loss
  • Pages updated: /dashboard, /dashboard/search, /dashboard/projects, /dashboard/alerts, /dashboard/errors, /dashboard/traces, /dashboard/security, /dashboard/security/incidents, /dashboard/admin/organizations, /dashboard/admin/users, /dashboard/admin/projects, /dashboard/settings/members
• Automated Helm chart releases: every stable Docker image release now triggers a repository_dispatch to logtide-dev/logtide-helm-chart, which auto-bumps appVersion and chart version (patch), commits, and publishes a new chart release to the Helm repo on GitHub Pages

Fixed

• API 400 responses now include a details array with field-level validation errors instead of just a generic message. Covers both Fastify/AJV schema validation and Zod validation errors (including uncaught ZodError that previously returned 500)
• Admin pages returned 502 Bad Gateway on direct load/reload: the admin layout ([email protected]) breaks out of the dashboard layout chain, so ssr = false was not inherited; added a dedicated +layout.ts to the admin section
• /dashboard/admin/projects/[id] crashed with “Something went wrong” due to formatDate being called but not defined (function was named formatTimestamp)
• POST /api/v1/logs/identifiers/batch slow: the route was calling reservoir.getByIds (hitting ClickHouse/TimescaleDB/MongoDB) only to verify project access, then querying log_identifiers (PostgreSQL) separately. Since log_identifiers already stores log_id → project_id + identifier data, the storage engine call is now bypassed entirely — one PostgreSQL query replaces the N×storage-engine-roundtrips loop. Added bloom filter skip index on id in ClickHouse and a standalone id index in TimescaleDB (migration 032) for getByIds used by findCorrelatedLogs
• GET /api/v1/logs/hostnames taking 8+ seconds: the 6h window cap was only applied when from was absent — explicit from params (e.g. 24h range from the search page) bypassed it and triggered a full-range metadata scan; cap now clamps any window to 6h max. Added limit: 500 to the distinct call. Per-engine optimizations: ClickHouse adds a hostname materialized column (computed at ingest, eliminates JSONExtractString at query time) and uses it directly in distinct queries; TimescaleDB adds a composite expression index (project_id, (metadata->>'hostname'), time) (migration 032); MongoDB adds a sparse compound index on metadata.hostname. All three engines also now extract the metadata field in a subquery (once per row vs 3×)