Header id generation in the notebook needs to be majorly reworked.

[mpacer]

Both in terms of generating invalid ids for headers, as well as in not adhering to the conventions already present in other header id mechanisms (e.g., pandoc), and in not using user generated ids, our current solution for generating ids is problematic.

If you want a quick overview of some of these issues see:

# Header:

I'm a markdown link to the [second header](#Second-header)!

## Second header {#Second-header}

##  3: Third ,#">header 

the ids it produces include:

#Header:
#Second-header-{#Second-header}
#3:-Third-,#">header

because all we're doing is replacing spaces ( ) with hyphens (-) and stripping leading and closing whitespace.

Pandoc in contrast generates these three header ids:

#header 
#Second-header
#third-header

related to some issues discussed in jupyter/nbconvert#469.

[takluyver]

• We can and should certainly fix where we generate invalid IDs. Do we know which characters are valid in IDs?
• It's probably reasonable to tidy up corner cases around punctuation and numbers in the process, which hopefully not too many people are relying on.
• I don't think we should try to follow pandoc on lowercasing all IDs, because it would break too many links in existing notebooks.

[mpacer]

I'm ambivalent about lowercasing everything and am prepared to defer to others on that.

What is likely going to require the most reworking is that we should be autoextracting and replacing ids with content in curly brackets containing a leading octothorpe ({#…}). But then we're going to need to validate that as well.

So it's probably going to have to go through two checks:

1. detect if there is a manually defined id target, if none autogenerate an id
2. validate the id (either manually or via autogeneration)