feat: built-in crash reporting, debug logging, and telemetry support

[zrosenbauer]

Description

Add first-class support for error/crash reporting, debug logging, and optional telemetry in kidd so CLI authors don't have to roll their own.

Motivation

When a CLI crashes or errors silently (especially on unfamiliar platforms like Windows), there's no easy way to:

• See what went wrong without reproducing with --verbose
• Collect crash data from users for debugging
• Track error patterns across the user base

Currently each CLI has to build its own debug logging, crash file writing, and error reporting - this should be a framework concern.

Proposed Features

1. Crash & Debug Log Writer

• Automatically write crash/error data to a well-known location (e.g. ~/<dotdir>/.debug/crashes/)
• Include stack traces, platform info, CLI version, command + args (sanitized)
• Users can share these files when reporting issues (<cli> debug --last to print the latest crash log)
• Configurable log retention (auto-prune old logs)

2. Telemetry / 3rd-Party Integration

• Optional telemetry hook that CLI authors can wire up to services like:
  • Statsig (feature flags + event logging)
  • Sentry (crash reporting)
  • PostHog (product analytics)
  • Custom HTTP endpoints
• Opt-in by default with clear disclosure, easy opt-out (<cli> telemetry off)
• Minimal API surface, e.g.:

  cli({
    telemetry: {
      provider: statsig({ key: '...' }),
      // or: provider: sentry({ dsn: '...' }),
      // or: provider: custom({ send: async (event) => { ... } }),
      consent: 'opt-in', // 'opt-in' | 'opt-out' | 'required'
    },
  })

3. Error Context Enrichment

• Automatically capture platform, arch, Node version, CLI version, shell, terminal info
• Attach to crash logs and telemetry events
• Respect privacy - no PII, no args that look like tokens/keys

Use Cases

• User reports "cli command fails silently on Windows" - we can ask them to run <cli> debug --last and share the crash log
• Proactively detect that Windows users are hitting a specific error pattern via telemetry
• Feature flag rollouts via Statsig to gate experimental features

Additional Context

This came up while debugging silent failures on Windows in Parallels - the CLI exited with no output and no crash log, making it impossible to diagnose without source access.