feat(ui): add bubbletea-backed spinner for long-running steps

[dipto0321]

Summary

Wraps bubbletea in a small Spinner interface and wires it into nodeup upgrade's long-running steps (manifest fetch, per-version snapshot, per-version install, set-default, global-package restore). Plain-mode callers get a one-shot "label ..." line per operation so CI / piped-output stays log-friendly; Fancy-mode callers get an animated braille-pattern spinner driven by a real tea.Program.

This is PR2 of 3 for #105 (PR1 was the scaffold via #74; PR3 will add huh prompts + detector.ResolveInteractive; PR4 will migrate the remaining call sites — check, list, packages, config — to ui.Writer).

Why a separate PR

The Charm-stack migration is large enough that mixing it into one PR would balloon review surface and confuse the commitlint history (PR4 is mostly cmd.Printf → ui.Out(w) mechanical substitutions; PR3 introduces interactive prompts). Splitting along the existing #74 phasing keeps each PR reviewable in ~10 minutes.

Implementation

internal/ui/spinner.go

• Spinner interface: Start() (stop func()) and Mode() Mode.
• NewSpinner(mode, label, out): routes to fancySpinner (FancyMode + non-nil writer) or plainSpinner (everything else).
• plainSpinner: prints "label ..." on Start(). Stop() is a no-op so the caller can emit the success/failure line via Writer.Success / Writer.Error without producing a second spinner record.
• fancySpinner: pipes a tea.Program through a transient stderr handle so spinner frames don't interleave with Writer.Success lines emitted during the same operation. Standard braille-pattern frame set, 80ms tick. Stop() calls p.Quit(), waits ≤200ms for teardown, closes the pipe.
• WaitWithSpinner(ctx, s, work): the production helper. Returns nil on success, work's error if work errors, or ctx.Err() if ctx fires first. Does not preempt work — callers who need preemptible work must thread ctx into work themselves.

internal/ui/spinner_test.go (new)

• TestNewSpinner_PlainModeHonored — plain prints label, Stop is no-op/idempotent.
• TestNewSpinner_FancyModeSelected — fancy mode decision + lifecycle safe under non-TTY go test.
• TestNewSpinner_FancyNilWriterDegradesToPlain — (FancyMode, nil) doesn't NPE inside tea.
• TestWaitWithSpinner_NilSpinnerPassesThrough — useful for tests that don't want a spinner.
• TestWaitWithSpinner_HappyPath — work nil → helper nil.
• TestWaitWithSpinner_WorkErrorReturned — work error surfaces.
• TestWaitWithSpinner_ContextCancelReturns — ctx cancel returns context.Canceled, work goroutine is not preempted (left parked, released by test's defer).

internal/cli/upgrade.go

• spinnerFor(cmd, label) constructs the spinner matching the active ui.Writer.Mode(), rendered to cmd.ErrOrStderr(). Stderr is deliberate: plain-mode users can pipe nodeup upgrade | jq and the human-side channel still has the spinner line.
• Five call sites wrapped in ui.WaitWithSpinner: node.FetchManifestCtx, per-version packages.Snapshot, per-version m.Install, m.SetDefault, packages.RestoreFromSnapshot.

go.mod / go.sum

• Adds github.com/charmbracelet/bubbletea v1.3.10. New transitive deps: charmbracelet/x/ansi v0.10.1, erikgeiser/coninput, mattn/go-localereader, muesli/ansi, muesli/cancelreader, golang.org/x/text v0.3.8, golang.org/x/sys v0.36.0. (The x/ansi bump from 0.8.0 → 0.10.1 is bubbletea's requirement; safe per upstream changelog.)

Edge cases I considered before opening

• TTY-faking in unit tests: bubbletea's renderer needs a real terminal to produce frames. Under go test stdin/stdout are pipes. Tests pin the decision point (FancyMode → spinner.Mode()==FancyMode) and the lifecycle (Start/Stop don't panic/deadlock) but don't assert on rendered frame bytes. End-to-end "does the spinner look right" is a manual smoke test, documented in PR3's body when huh lands.
• Nil writer under FancyMode: tea.NewProgram(nil-writer) would NPE deep in bubbletea. We degrade to no-op rather than crashing mid-upgrade. spinner_test.go pins this.
• Double-prompt cleanup interaction: the cleanup-prompt goroutine still uses bufio.NewReader(cmd.InOrStdin()). In FancyMode the spinner was rendering to stderr; stdin stays free for the prompt. Verified by reading the existing runCleanupPrompt codepath and confirming it never touches stderr.
• Stop() called twice: plain spinner's Stop is a no-op (idempotent). Fancy spinner's Stop calls p.Quit() then waits; a second call would p.Quit() again (also idempotent in bubbletea), close already-closed pipes (errors swallowed), and return. No goroutine leak in either case.
• Plain-mode log count: per the project's V2 invariant, one log record per operation. Plain spinner's "label ..." line is the only spinner-side record; the success/failure line comes from the caller's Writer.Success / Writer.Error. Tests pin that Stop() writes no extra bytes in plain mode.
• tea.WithoutSignalHandler: without this, bubbletea would install its own SIGINT handler that races with our cmd.Context() cancellation. We install our own ctx-aware path and pass WithoutSignalHandler so bubbletea leaves signal handling alone.
• tea.WithoutCatchPanics: bubbletea wraps every Update call in a recover-by-default. We opt out so a panic in the spinner model surfaces as a real panic (test runner / Go runtime catches it) instead of being silently swallowed and the program just stopping — which would manifest as a hung upgrade.

Verification

$ make ci
go mod tidy
gofmt -s -w .
go vet ./...
golangci-lint run ./...
0 issues.
go test -race -coverprofile=coverage.out -covermode=atomic ./...
ok  	github.com/dipto0321/nodeup/internal/cli	2.345s	coverage: 37.3%
ok  	github.com/dipto0321/nodeup/internal/ui	1.776s	coverage: 73.1%
...
total: 62.4%

Plus a manual make build && ./bin/nodeup upgrade --help to confirm the binary still launches cleanly.

Out of scope (deferred to PR3 / PR4 per #105 phasing)

• huh-backed prompts for cleanup and the manager-resolution decision (PR3 — also adds detector.ResolveInteractive).
• Migrating cmd.Printf in check, list, packages, config, and the remaining upgrade lines (PR4). The plain-mode spinner line emitted by ui.NewSpinner here is already ui-routed, so PR4 just has to swap the cmd.Printf around the spinners, not inside them.

Part of #105. PR3 (huh prompts + ResolveInteractive) and PR4 (full call-site migration) will follow in this same issue. This PR alone does not close #105.

[cocogitto-bot]

✔️ 19ed51e - Conventional commits check succeeded.