Run all archived messages when embedded widget is uninitialized

[Yoshanuikabundi]

This PR implements the proposal discussed in #1063.

I made two changes that weren't discussed there:

1. _ngl_is_initialized is set in createStage, not handleEmbed, to ensure it is always set when the JS runs (but also after handleEmbed)
2. I had to make on_msg async. This is because it is essential that each message completes before the next one is triggered to avoid things like trying to add representations to a missing component. I learnt this the hard way. My understanding is that this won't change behavior - on paths where on_msg doesn't await, execution will never be suspended and thus the function will run synchronously; on the other hand, when the function does await it's the last thing it does, so the only difference is that the on_msg function sticks around on the queue instead of the already async inner function. What it does do is allow on_msg to be awaited so that messages can be executed and completed in order. This may be of use elsewhere, I'm not sure - I feel like NGLView does sometimes "forget" something I told it.

I also added the jupyterlab~=3.0 pin to the Conda environment as discussed.

I have a lot of files that have been generated by NPM (or something), including three that are already checked in - so I've added them in a separate commit in case this is a mistake of some sort.

	modified:   nglview/static/index.js
	modified:   nglview/static/index.js.map
	modified:   nglview/staticlab/package.json

Hope this is useful! I think from what I've learned I can hack something together for my cookbook even if this doesn't get merged, so thanks for your help!

Closes #1063

[hai-schrodinger]

Does this PR actually work as expected?

[Yoshanuikabundi]

Yep, seems to! I was surprised too 😂 I made a notebook that creates a widget, clears it's representations and adds new ones. It appears correctly when executed either with jupyter nbconvert --execute --to html notebook.ipynb or in JupyterLab. I did have to manually copy over the compiled JavaScript file but my understanding is thats just how the whole embedded widget packaging thing works. I'll put together a more comprehensive demo today.

[Yoshanuikabundi]

Sorry Hai, I'm not going to have time to put that more comprehensive demonstration together this week. Here's the test I was developing off of, along with the compiled JS and the converted HTML file - let me know what more you need to review this!

pr-demo.zip

[hainm]

should update await for other places using on_msg(msg) too?

[Yoshanuikabundi]

Uhmmm possibly. I didn't make this sort of change to avoid changing behaviour. The other places calling on_msg are:

l450 (in handleEmbed()):

        // fire any msg with "fire_embed"
        that.model.get("_ngl_msg_archive").forEach(function(msg){
            if (msg.fire_embed){
                that.on_msg(msg);
            }
        })

This should probably be awaited - it's in an async function already and its possible that _set_representation_from_repr_dict could try to set a representation on a component added here? The comment above _set_representation_from_repr_dict seems to suggest that completing all these messages is a requirement, and completing them out of order could mean assigning the wrong representations. (I will make this change shortly)

l152 (in handleMessage):

        this.model.on("msg:custom", function(msg){
           this.on_msg(msg);
        }, this);

From what I can google, ipywidgets/backbone don't seem to treat async callbacks any different to regular callbacks, so making the callback function async and adding an await here wouldn't do anything.

[hainm]

got it. thanks.

[hainm]

should we check for ngl_stage_params instead?

in another line:
var ngl_stage_params = that.model.get('_ngl_full_stage_parameters');

Basically if the ngl is not initialize, check for empty of something?

[hainm]

[hainm]

or check for both ngl_full_stage_parameters and _camera_orientation?

[Yoshanuikabundi]

Done! It looks _ngl_full_stage_parameters object is already modified in createStage, so I've just made this a shallow copy so that changes aren't reflected in the shared state until we deliberately sync them back up. I think this is more robust to future changes than checking if _ngl_full_stage_parameters is the value it is modified to.

[hainm]

I have a lot of files that have been generated by NPM (or something), including three that are already checked in - so I've added them in a separate commit in case this is a mistake of some sort.

	modified:   nglview/static/index.js
	modified:   nglview/static/index.js.map
	modified:   nglview/staticlab/package.json

yeah, those files are intentional to be included in nglview folder so we don't need npm to install the dev nglview python package.

[Yoshanuikabundi]

OK I've addressed those points! One thing I've noticed is that js/package.json still thinks its 3.0.1 - does that need to be updated?

[hainm]

One thing I've noticed is that js/package.json still thinks its 3.0.1 - does that need to be updated?

I will do when doing release.

[hainm]

Thanks @Yoshanuikabundi very much for your contribution.