Dashboard: share_url() without data gives a URL whose updates fail #143

Problem

share_url() on a Dashboard without data returns a URL that opens but never shows data. Its docstring says the method neither starts a server nor registers the source, and that the server must already serve source_name. That reads as "no data needed", but data is needed.

import polars as pl
from flexviz import Dashboard

lf = pl.LazyFrame({"t": [0, 1, 2], "value": [1.0, 2.0, 3.0]})
for label, d in [("Dashboard()", Dashboard()), ("Dashboard(lf)", Dashboard(lf))]:
    d.add_figure().add_line(x="t", y="value")
    print(label, d.to_spec(source_name="logs").figures[0].source)
# Dashboard() None
# Dashboard(lf) logs

The cause is Dashboard.to_spec(): src = source_name if self._backend_lf is not None else None. In a browser, GET /view returns 200. Then each /dashboard/update returns 500 (KeyError: 't' in trace/line.py, get_aggregation_spec, for a line), and the console logs "initial load failed, panels are empty". Nothing fails in Python.

It is easy to hit in web apps (#73, #141). The page script builds the URL, and the server registers the data in another file or process. The web apps guide now works around it: build the Dashboard on a lazy scan of the same file.

Proposal

Honour source_name in to_spec(), with or without data. This changes share_url(), save_spec(), and to_spec() together. The server already checks axis-link types against the registered schema on /view and on each update (_check_link_types), so the client-side link check can stay limited to dashboards with data. show() registers no source for a dashboard without data, so it passes no source_name and keeps its behaviour.

The alternative is to raise ValueError in share_url() when the dashboard has no data. That is smaller, but it keeps the second scan in web apps for no reason.

Done when

• Dashboard().share_url(source_name="x") gives figures with source == "x", and the page loads data from a server that serves x.
• A test covers it.
• The web apps guide drops the "same data" rule, or limits it to dashboards with linked axes.