Allow Blocks instances to be used like a Block in other Blocks

[abidlabs]

This PR does a number of things:

• Handle Markdown better: cleans up indentation from markdown blocks so that if every line is indented a certain amount, that is removed uniformly. Before this, any indented Markdown would show up as codeblocks (try running demo/xray_blocks/run.py to see the difference)
• Allow running multiple Gradio servers from one Python session / jupyter notebook / colab notebook
• Allow using Blocks/Interfaces in other Blocks:

So that we can do things like this:

with gr.Blocks() as bl:
    with gr.Tabs():
        with gr.TabItem("Text"):
            gr.Interface(lambda x:x, "textbox", "textbox")
        with gr.TabItem("Image"):
            gr.Interface(lambda x:x, "image", "image")
    bl.launch()

• Creates the TabbedInterface abstraction to make the above easier. Just a rough draft, feedback on naming and API welcome here

io = gr.Interface(lambda x:x, "textbox", "textbox")
io2 = gr.Interface(lambda x:x, "image", "image")            

t = gr.TabbedInterface([io, io2])
t.launch()

[aliabid94]

This seems not the right way to do this. Why can't you just call interface.render() instead of creating a new Interface?
Also I'm a little skeptical we should even create TabbedInterface, instead of allowing users to naturally explore Blocks as they want more customizability.

[abidlabs]

Why can't you just call interface.render()

There's a few Interface kwargs that don't make sense to have within this context. For example, theme. So I just decided to pick the kwargs that make sense to include here.

Also I'm a little skeptical we should even create TabbedInterface, instead of allowing users to naturally explore Blocks as they want more customizability

Open to discussion about this. We should keep in mind that we've heard directly from several users that they want something like this rather than having to delve into the lower-level Blocks API. E.g. #690 (comment)

[aliabid94]

is it possible that instead of using "without_rendering", it just checks to see if its within a BlockContext and only calls render if that's the case?

[abidlabs]

That doesn't work. Consider the following example:

with gr.Blocks() as bl:
    with gr.Tabs():
        with gr.TabItem("Text"):
            gr.Interface(lambda x:x, "textbox", "textbox")
        with gr.TabItem("Image"):
            gr.Interface(lambda x:x, "image", "image")
    bl.launch()

Inside the Interface class, you create Component objects once, and then render them later.

Create components:

gradio/gradio/interface.py

Line 182 in 5c44ae3

self.input_components = [get_component_instance(i) for i in inputs]

Render:

gradio/gradio/interface.py

Line 496 in 5c44ae3

component.render()

In this case, both statements are inside a BlockContext (the outer level Blocks bl), so the component would end up getting rendered twice. That's why we need to explicitly escape the context.

[abidlabs]

I'm going to remove TabbedInterface and will instead create a separate PR for these types of Interfaces. That'll be cleaner.

[aliabid94]

Lgtm