feat(desktop): verbose logging toggle + Sentry/PostHog instrumentation expansion

[richiemcilroy]

What

Two complementary changes that make support diagnoses dramatically easier:

1. Verbose logging toggle (General Settings → Diagnostics)

Adds a Diagnostics card to General Settings with three controls:

• Verbose logging toggle — flips the global tracing_subscriber level filter to TRACE at runtime via a reload::Layer, so every span/event from the recording, rendering, export, and upload pipelines is captured to the rolling log file. Persisted across restarts in GeneralSettingsStore.
• Reveal logs — opens the platform-native logs folder (~/Library/Logs/so.cap.desktop on macOS, %LOCALAPPDATA%/so.cap.desktop/logs on Windows) via tauri-plugin-opener.
• Upload logs — co-locates the existing upload_logs flow next to the toggle so users can ship the captured trace logs in one click after reproducing.

Implemented via tracing_subscriber::reload::Layer<LevelFilter, _> in main.rs, a small VerboseLoggingHandle managed in Tauri state, and two new set_verbose_logging / open_logs_folder commands.

2. Always-on telemetry coverage (no toggle required)

So we don't always need users to flip the toggle and re-upload to diagnose issues, the Sentry/PostHog surface is now far richer by default:

• sentry-tracing layer — every error!() event becomes a Sentry event, and every warn!() / info!() rides along as a breadcrumb. The desktop app already has hundreds of structured tracing calls across recording, encoder, muxer, upload, camera, mic, GPU, and editor flows; they now all flow into Sentry automatically with their fields and span context attached.
• Sentry scope tags at startup — os, arch, cap_version, instance_id, cap_backend (cloud vs self_hosted), verbose_logging state. This lets us filter issues per environment.
• Export failure context — with_scope adds export.format, export.fps, export.cursor_only, export.force_ffmpeg, and project_path to the captured Sentry message instead of just the bare error string.
• GPU init failures — wgpu::request_adapter / request_device errors that previously dropped silently via .ok()? are now logged with error! (adapter_name, adapter_backend, adapter_device_type, error) so they reach Sentry.
• multipart_upload_failed PostHog event gains stage (video_join / instant_multipart / studio_segment), retried_chunk_count, and bytes_uploaded_mb so we can tell which upload path is breaking and how far it got.
• recording_started PostHog event now sends real target_width / target_height derived from the capture target's display / area bounds (was hardcoded to 0).

Walkthrough

Manual UI testing was not run — this cloud agent is Linux, but the desktop crate (scap-targets, whisper-rs-sys, cidre, …) only builds on macOS / Windows. The Rust changes pass cargo check for sentry-tracing resolution and were carefully matched against existing patterns:

• The new tracing → Sentry layer is the standard sentry::integrations::tracing::layer() with the recommended EventFilter mapping.
• Scope tags reuse the same sentry::configure_scope block that already runs at startup.
• Export tags are added via sentry::with_scope so they apply only to the captured failure.
• request_device already returns Result<(Device, Queue), Error> in wgpu 25 (matches existing usage in screenshot_editor.rs).

The desktop CI matrix on macOS + Windows runners exercises the full clippy -D warnings gate.

Notes

• Cargo.toml adds the tracing feature on sentry. Cargo.lock updated.
• tauri.ts is regenerated by tauri-specta but tracked in this repo (see prior chore(desktop): regenerate tauri bindings ... commits). Updated by hand in lockstep with the new commands and GeneralSettingsStore field — the next debug run will produce the same output.
• No behavioural change to upload retry logic or to the enable_telemetry opt-out: all PostHog events still respect the existing telemetry toggle, and Sentry is only initialised when CAP_DESKTOP_SENTRY_URL is set at compile time (release builds only, debug builds drop events via before_send).