Fix: retry button on update failures, and stop the dialog redrawing per chunk

[broisnischal]

Reported from the update dialog: no way to retry after a failed check, and both checking and updating feel slow.

Retry

A failed check printed the raw transport error and stopped there. The only recourse was to dismiss the dialog and find the command palette entry again. There is now a Retry button next to Not now, and it re-runs whatever actually failed: the download when an update is already in hand, otherwise a fresh check.

Header said the wrong thing

The screenshot shows Checking for updates… above an error body. The header tested checking before status, and checking is still true while the failed check unwinds, so the spinner text outranked the result. A settled outcome now wins, and starting a new check clears the previous one so the header is not stale in the other direction either.

The endpoint URL is a single unbreakable token and ran past the dialog edge; it wraps now.

Speed

	before	after
check() timeout	none (HTTP stack default)	15s
dialog redraws while downloading	one per HTTP chunk, thousands for a 100MB build	one per frame

downloadAndInstall reports a Progress event per chunk and every one of them wrote $state, re-rendering the dialog at that rate. The running total is now accumulated outside reactivity and published on a requestAnimationFrame.

The download is deliberately left unbounded. A timeout there would break exactly the slow connections that need it most; only the small latest.json check is bounded.

Verification

Forced the reported failure and rendered the dev build headlessly:

Couldn't check for updates
Stroke could not reach the update server.
error sending request for url (https://github.com/stroke-app/stroke/releases/latest/downl…
[ Not now ]  [ ⟳ Retry ]

Header, wrapped error text and both buttons confirmed in the DOM.