Refactor for maintainability: split TUI and centralize health hints

[pol-cova]

Summary

Applies the maintainability recommendations from codebase review: smaller focused files, centralized threshold logic, dead code removal, and expanded test coverage.

Changes

Dead code removal

• Delete unused internal/metrics/types.go (Panel, TimeSeries were never referenced)
• Remove unused config fields theme and prometheus_presets (presets remain in prometheus.Presets)
• Fix config test to use valid panel names (processes, docker instead of non-functional overview)

Centralized health & thresholds

• Use config.Thresholds as the single threshold type via alias in health
• Add health.ApplyDefaults() and health.HealthyMessage constant
• Refactor observe ask to use health.HintsWithThresholds and respect observe.yaml thresholds
• Reuse prometheus.PresetByName for 5xx query text in assistant

TUI split (616-line monolith → 4 files)

• dashboard.go (239 lines) — Bubble Tea lifecycle, collection, input
• render.go (173 lines) — bars, sparks, charts, process tables
• views.go (174 lines) — overview, chart, integration, help, detail views
• styles.go (12 lines) — lipgloss styles

Collector refactor

• Split local.Collect() into focused helpers (collectCPUAndMemory, collectNetworkRates, etc.)
• Reuse processSummary in topProcesses

Performance / clarity

• Use history.ReadSamples() for hints instead of copying 1800 snapshots every tick
• Discover Prometheus metrics once at startup
• Name constants: collectEveryNTicks, rename collector interface to metricsCollector

Tests

• assistant_test.go — health hint integration, config thresholds, prometheus preset
• health/defaults_test.go — threshold defaults
• prometheus/preset_test.go — preset lookup
• metrics/remote/remote_test.go — snapshot decode and SSH error handling
• Expanded docker_test.go for unit parsing

Coverage: 36.6% → 39.6% overall; assistant 0% → 91%, remote 0% → 60%

Verification

gofmt -w .
go vet ./...
go test ./...
go build .

Skipped (follow-up)

• Async metric collection (goroutine + channel) — larger behavioral change
• Docker polling interval throttle — needs UX decision
• topProcesses / net.Connections optimization on busy hosts — performance project
• README update for valid panels config values