fix: honor context cancellation in node health/exit wait loops

[mvanhorn]

Summary

Make the per-node deploy wait loops honor context cancellation so a stuck containerlab deploy unwinds on Ctrl-C instead of hanging until SIGQUIT.

Why this matters

When a node container crashes or disappears mid-deploy, a dependent worker in scheduleNodeWorkerF (core/clab.go) spins on time.Sleep(time.Second) waiting for a state that never arrives. Ctrl-C only cancels the deploy context, so these bare sleeps and the blocking stage waits never observe the cancellation and the process hangs with open sockets until SIGQUIT (issue #3162; the current workaround is --skip-post-deploy).

The fix closes the hang at every layer it could occur:

• Replace the bare time.Sleep calls in the WaitForHealthy and WaitForExit poll loops, and the StartupDelay sleep, with select { case <-ctx.Done(): return; case <-time.After(...): } (core/clab.go).
• Make DependencyNode.EnterStage wait on its stage waitgroup through a select on ctx.Done() so a node blocked on an unmet dependency returns on cancellation instead of blocking forever (core/dependency_manager/dependency_node.go). This is the single blocking wait path for all stage dependencies, so making it cancellation-aware unwinds dependent workers too.
• After each EnterStage call in the worker, return early when ctx.Err() != nil. In scheduleNodes, move wfcwg.Done() into a defer, bail on cancellation after the create-stage wait, and make the workerChan send a select on ctx.Done() so the channel always closes and the send cannot block once workers have unwound.
• In Deploy, check ctx.Err() right after NodesWg.Wait() and return the cancellation error instead of running host endpoint deployment, vxlan stitching, exports, and inventories on a partially deployed lab (core/deploy.go).

Happy-path behavior is unchanged: the loops still poll every second and break when a node turns healthy or exits, and post-deploy ordering is untouched.

Testing

Added focused unit tests: core/clab_test.go drives scheduleNodeWorkerF into the health wait loop with a context cancelled during the health check and asserts the worker returns rather than blocking past a short deadline; core/dependency_manager/dependency_node_test.go blocks EnterStage on an unmet dependency, cancels the context, and asserts it returns. gofmt is clean and core/ plus core/dependency_manager/ (and their test binaries) build and vet under GOOS=linux. Full go test execution requires a Linux host because containerlab pulls Linux-only dependencies.

Fixes #3162