Failed auto-saves after AI apply aren't logged/reported anywhere

Depends on #4848 Split out from #4876 phase 2.

The gap

When the AI assistant applies a workflow to a brand-new canvas, we
auto-save it (saveNewWorkflow in useAIWorkflowApplications.ts). If
that save fails, the only trace of it is a console.error in the
browser — nothing is reported anywhere we'd actually see it (no Sentry
event, no server-side log).

The user still gets a toast, so this isn't a silent failure from their
side. But we have no visibility into how often this happens or why —
if auto-save is flaky for some users we won't know unless they tell us.

Why it's not already covered

Most workflow-save failures that reach the server are already logged
there (and forwarded to Sentry via the existing Logger handler) — see
Session.handle_call({:save_workflow, ...}). The gap is specifically
failures that happen client-side and never reach that code path (e.g.
a channel push timeout, or the client being disconnected at the moment
the auto-save fires).

Frontend Sentry (@sentry/react, assets/js/app.js) is currently
disabled boilerplate — never wired up for real capture — so it's not a
ready-made fix here.

What to add

Some way for this client-side failure to surface as telemetry we'd
notice — either by reporting it back to the server to log (reusing the
Logger → Sentry pipeline that already works for other save failures),
or another mechanism if there's a better fit. Low complexity, no new
infra needed either way.

Start in useAIWorkflowApplications.ts at saveNewWorkflow, then trace Session.handle_call({:save_workflow, ...}) and the existing Logger handler. Compare the client-side failure path with the server-side logging path and choose how telemetry should be surfaced. Done means failed auto-saves produce observable telemetry without adding infrastructure.