fix(web): clone subworkflowContexts tree in processEvent (#307)

[Jason Robert (jrob5756)]

Summary

Fixes#307: an agent node adjacent to a type: workflow sub-workflow step
could render stuck showing "running" even though the underlying store data
was already completed — a pure rendering/reactivity desync, not an engine
data bug (confirmed correct via the detail panel per the issue report).

Root cause

processEvent in workflow-store.ts clones nodes/groupProgress/
eventLog/activityLog on every event so React/Zustand reference-equality
checks pick up changes (per replaceNode's own doc comment: "Create a new
reference for a node to ensure React/ReactFlow detects the change") — but
it never cloned subworkflowContexts. Handlers reached through
activeTarget()/resolveContext()/resolveSlotPath() (subworkflow_started,
subworkflow_completed, agent events carrying a subworkflow_path, etc.)
mutate nested ctx.nodes/groupProgress/routes/highlightedEdges in
place, and subworkflow_started pushes onto subworkflowContexts/
children arrays in place. Nothing forced a fresh reference anywhere in
that tree, so any selector/useMemo keyed on subworkflowContexts (e.g.
useViewedNodes/useViewedGraphData in use-viewed-context.ts) could miss
an update depending on render/batching timing.

Fix

• Added cloneSubworkflowContexts(): a structural clone of the
  SubworkflowContext tree (each context plus its own nodes/
  groupProgress/routes/highlightedEdges/children/etc.), wired into
  processEvent alongside the existing clones.
• Audited replayState: it already rebuilds subworkflowContexts fresh
  from {} per replay and commits once at the end, so it needed no change.

Testing

There were previously zero frontend automated tests. Added a minimal
Vitest setup (vitest devDependency, vitest.config.ts, npm run test,
and a make test-frontend target) plus a regression test in
workflow-store.test.ts that:

• Simulates the reported repro (an agent looping back once, then an
  adjacent subworkflow_started/subworkflow_completed pair) and asserts
  the agent's node stays completed and subworkflowContexts gets a fresh
  reference on every mutating event.
• Covers a second case: a node update nested inside a running child
  sub-workflow context (via subworkflow_path), asserting the same
  reference-identity behavior.
• Verified fails without the fix: temporarily reverted the
  cloneSubworkflowContexts call and confirmed both tests fail with the
  exact Object.is reference-equality assertion; restored the fix and
  confirmed both pass again.

make test-frontend # 2 passed
npx tsc -b # clean
uv run ruff check src tests # All checks passed!
uv run ty check src # 1 pre-existing warning (dialog_evaluator.py), unrelated

Closes#307.

[Jason Robert (jrob5756)]

Ran a multi-agent code review (code-reviewer, pr-test-analyzer, comment-analyzer) against the original commit and addressed the findings in a follow-up commit:

• Performance/re-render fix: replaced the whole-tree clone of subworkflowContexts with a path-scoped clone-on-write (resolveMutableContext/resolveMutableSlotPath). Only the mutated branch's ancestor chain gets a fresh reference now; untouched siblings (and root-level events with no active sub-workflow) keep their existing reference, so this no longer forces every subworkflowContexts-dependent component to re-render on unrelated events.
• New regression tests: nested (grandchild) sub-workflow propagation, concurrent for_each siblings not cross-cloning each other, root-only events never touching subworkflowContexts, and field preservation across a clone-and-mutate cycle. Verified each fails when the corresponding part of the fix is reverted.
• CI: added a frontend job (typecheck + Vitest + build) so this regression suite runs on every PR instead of relying on someone remembering make test-frontend.
• Comment hygiene: trimmed doc-comment cross-references prone to drift (pointing at another comment, naming specific hook consumers, citing the issue number as load-bearing).

All 6 frontend tests pass, tsc -b is clean, and ruff check / ty check are unaffected.