feat(logs-page): add a Logs page for browsing and fetching vehicle logs

[dakejahl]

Summary

Adds a Logs page listing every log logloader knows about — on the vehicle, downloaded, uploaded — and lets the operator pick which ones to fetch and publish. Uploaded logs link straight to the plot in the onboard Flight Review.

Depends on ARK-Electronics/logloader#32, which the submodule bump points at; merge that first.

Problem

logloader decided on its own what to download and upload, and there was no way to see what it had or to ask for a particular log. Its counterpart change makes fetching request-driven, which needs somewhere to make requests from.

Two things surfaced while wiring it up that are not specific to this page. The dev server gzips responses by default and its compression middleware buffers until it has enough bytes, which for an SSE stream is never — measured through npm run serve with a browser's Accept-Encoding: 0 bytes and an EventSource stuck at readyState 0, against 3332 bytes through the gateway. That has been silently breaking the journal viewer, the network-stats stream and the firmware-progress stream in dev too. Separately, http-proxy-middleware v3 reads handlers from on, so the v2 onError/logLevel options every proxy block uses are dead: a stopped upstream answers with the default plain-text 504 rather than the JSON the UI parses.

Upgrading an existing install was also unsafe. merge_configs.py takes its structure from the new template and prunes scalars the template lacks — right for a removed field, wrong for a renamed one. Simulating an upgrade of a real logloader.toml produced [upload_remote] enabled = false and the stock url, so a vehicle configured to upload to a remote Flight Review would silently stop.

Solution

logloader serves a localhost API on :3005; the gateway proxies it at /api/logloader, including the SSE stream that drives live transfer progress. The page follows the house patterns: Options API, an axios service module, an inline EventSource with named events, and the SystemPage loading/error/retry triad.

Select all or a subset, then Download & Upload, Download Only, Upload, Cancel — or Refresh: logloader no longer polls the vehicle at all, so the page asks for a fresh listing when it loads and on demand, and a "Logger running" chip mirrors MAV_SYS_STATUS_LOGGING so it is visible why an automatic fetch is waiting. A log the vehicle is still writing never appears, so it cannot be selected at a size it will not keep. An upload records only the relative path Flight Review redirects to, so targets carry their base url and a remote plot link resolves against the right origin rather than the ARK-OS host.

compress: false on the dev server, on: { error } for all five proxies, and migrate_logloader_config.py run from postinst before the merge so a rename reaches merge_configs as an ordinary same-shape merge.

Review of the wiring found three more, all fixed here: supplying on.error in http-proxy-middleware v3 replaces its error plugin and its guards, so an upstream dying mid-SSE made writeHead throw on a response whose headers were long sent — an uncaught exception in a socket callback that took the whole gateway down; EventSource only retries network failures, so the gateway's 502 while the daemon restarts closed the stream for good and the page said "reconnecting" forever; and the migrated download.use_burst was absent from the new config template, so merge_configs pruned the very key the migration had just preserved.

logloader's config tables are one level deep because the shared TomlEditor renders exactly one level; with two, every upload setting silently vanished from the Services page config editor. The editor now says so rather than rendering nothing.

Verified against PX4 SITL, a local flight-review, and on hardware — a Jetson running this branch against an ARK FMU v6X on PX4 1.18, upgrading a real pre-FTP install: first launch queues only the newest log, a log closing mid-run is picked up automatically (logger cycles via nsh, end to end to a plot link in the onboard Flight Review), select-all fetches and publishes the rest, a restart re-downloads and re-uploads nothing, delete/re-fetch and cancel behave through the gateway, the SSE stream flows unbuffered through nginx and the dev server, and the pre-FTP database plus old-format config carries its state and settings across.