…del (#6868)

* feat(workflows): wire workflow step lifecycle events in the paywall

Port of RevenueCat/purchases-android#3487

Emits WorkflowEvent.stepStarted / stepCompleted during multi-step workflow
paywalls, building on the model and wire format added in #6858.

- Adds Purchases.track(workflowEvent:) and routes it through the existing
  FeatureEvent path, exposed via PaywallPurchasesType and PurchaseHandler
- Adds WorkflowStepEventTracker: builds events with a per-impression traceId,
  entry reasons (start/forward/back), and terminal-step detection
- Wires WorkflowPaywallView to emit at four points: initial step, forward,
  back, and terminal completion (anchored to onDisappear like paywall_close)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(workflows): don't emit terminal stepCompleted when no page rendered

trackTerminalCompletionIfNeeded fired on dismiss whenever navigator.currentStep
was non-nil, even if no page ever rendered. That emitted a stepCompleted with no
preceding stepStarted in two cases: the initial step failing to build, and a
forward destination failing to build (the navigator advances but the page does not).

Gate terminal completion on transitionState.currentPage != nil, the same render
guard trackInitialStepIfNeeded already uses. This mirrors Android keying terminal
completion off _workflowState.value?.currentStepId, which is null when a step fails
to render. Extracted the decision into a pure shouldTrackTerminalCompletion helper
and unit tested it.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* test(workflows): extract WorkflowStepEventCoordinator for testable emission

The step event emission state (trace id, fire-once flags, "only if a page
rendered" gating) lived inline in WorkflowPaywallView, so the emission
sequence could only be verified manually. Android tests the equivalent logic
at the PaywallViewModel level; this brings iOS to parity.

Move that state machine into WorkflowStepEventCoordinator (composing the
existing WorkflowStepEventTracker). The view holds it as @State created in
init, so a new presentation yields a fresh traceId, and its four lifecycle
and navigation hooks become one-line delegations. Behavior is unchanged: the
currentPage gate, fire-once semantics, and forward/back/error/terminal
emission points all match the prior inline logic.

Add WorkflowStepEventCoordinatorTests covering the full appear/forward/back/
dismiss sequence, fire-once and no-page gating, traceId continuity and
per-impression reset, and a journey asserted at the MockPurchases
track(workflowEvent:) boundary (content/count, since the dispatcher does not
guarantee delivery order). The three bool-helper terminal tests in
WorkflowPaywallViewTests are migrated here.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(workflows): forward workflow events through MockPurchases.map wrappers

The map(purchase:restore:) and map(trackEvent:) extensions forwarded the
workflow fetch block but not trackWorkflowEventBlock, so a mapped copy's
track(workflowEvent:) silently dropped events. This is DEBUG-only (MockPurchases
is #if DEBUG) and reachable via PurchaseHandler.cancelling()/.with(delay:), so
production analytics were unaffected, but it could drop events in the
PaywallsTester debug-log path used to verify this feature.

Forward trackWorkflowEventBlock to the original, mirroring how workflowBlock and
the paywall trackEvent block are already forwarded. Flagged by Cursor Bugbot.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* refactor(workflows): overload track(_:) for WorkflowEvent

Rename trackWorkflow/trackWorkflowEvent to track(_:), overloading by event
type alongside the existing track(_ PaywallEvent) methods. Addresses review
feedback on PaywallEventTracker and PurchaseHandler.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>