<https://github.com/jgm/commonmark.js>

Caught by sjs!  This also improves performance measurably.

To prevent intra-word emphasis, we used to check to see if
the delimiter was followed/preceded by an ASCII alphanumeric.

We now do something more elegant:  whereas an opening `*` must
be left-flanking, an opening `_` must be left-flanking *and not
right-flanking*.  And so on for the other cases.

All the original tests passed except some tests with Russian
text with internal `_`, which formerly created emphasis but no
longer do with the new rule.  These tests have been adjusted.
A few new test cases have been added to illustrate the rule.

The C and JS implementations have both been updated.

Everything else gets a name starting with an underscore
and will be considered private.

This will allow us to keep the API stable while changing the
underlying data structure.  And it will avoid exposing properties
that have only an instrumental value in parsing.

Use these instead of direct property access.

This is the first step in a general move towards an API
like the one libcmark has.  This will allow us to have a stable
API that is independent of details of the AST.

This does a proper unicode case fold instead of just using
toUpperCase.  It is also faster, partly because we can do
one pass for space and case normalization.

Modified from the NPM package fold-case; proper credit given
in source and COPYING.

This way we can change it without changing four separate places
in the code.

This is used in both blocks.js and inlines.js.