🚦 refactor(cmd): app lifecycle struct for runTUI — replace dual WaitGroups + comment-ordered defers; ctx-aware watcher.Run

[martinciu]

🔍 Problem

runTUI (cmd/ccpulse/main.go:258-384, ~127 lines) wires everything inline — env resolution, devlog, cache open/rebuild, pricing, auto-recost, ingester, watcher, quota seam, TUI model — and then manages goroutine lifecycle by hand:

• Two separate sync.WaitGroups (bg, bfDone), two cancel funcs
• A five-deep defer stack whose relative order is load-bearing and documented only in a 16-line block comment (main.go:322-337)
• Adding any new background worker requires re-deriving the ordering contract from prose; shutdown_test.go exists precisely because this is fragile

Related watcher shape: pkg/watcher/watcher.go:56-81 — Run takes no context.Context (shutdown is solely via Close() closing fsnotify channels), and fsnotify errors are silently dropped (watcher.go:74-79). The watcher callback closes over the root ctx for ProcessFile, but the watcher itself can't be cancelled by that ctx.

The current design is correct (per #52's shutdown-discipline work) — this is a maintainability refactor, not a bug fix.

🛠️ Suggested shape

• An app struct holding the wired deps with start(ctx) / shutdown() methods, so registration order and teardown order live in code (a slice of stop funcs or errgroup) rather than in a comment
• errgroup.WithContext per lifecycle stage (backfill group + long-running group), waited in one deterministic teardown
• Add ctx to watcher.Run (select on ctx.Done()), giving the fsnotify error stream a propagation path; keep w.Close() as the post-p.Run() trigger

⚠️ Notes

• Shutdown-ordering bugs are subtle; safety nets: shutdown_test.go + goleak guards in pkg/tui and cmd/ccpulse (remember the bubbletea Tick.func1 goleak ignore applies to both guards).
• Reasonable to defer until the next time this code must change anyway.
• Related: 🚦 Shutdown discipline: signals, goroutine joins, no use-after-close #52 (original shutdown discipline).