feat: better implementation of initial scan and render

[Charlie-XIAO]

Extracted from #370.

Previously the manager simply waits for a few seconds before the initial scan and render, because we are unsure whether the canvas is ready to listen to the render event. The case can be summarized as this:

• If the canvas listener is ready first, then everything is okay.
• If the event is emitted before the canvas listener is ready, then the event will be missed.

This PR implements a mechamism that can be summarized as follows:

• The backend keeps a state that records (1) whether the canvas listener is ready and (2) a potentially pending message payload.
• When the canvas listener is ready, mark it in that state, and if there is a pending message payload there immediately emit to the canvas.
• For the manager, instead of just emitting, it calls a backend command that checks whether the canvas listener is ready first. If it is ready, immediately emit, otherwise it puts the payload in that state.

Some other changes:

• Previously the render event is per widget. Now it is for an array of widgets.