Previously, the processing rules do not allow nesting an element inside
of itself: ‹ ☞︎nested ☞︎tags☜︎ like ☞︎this☜︎☜︎ › do not work as expected.
This design choice was to (at least) appropriately handle ‹ ☞︎⟨this☜︎
weird ⟨case☜︎⟩ ›, where each inline should stop at the first delimiter.
Correct processing starts by looking for end sigils first, not start
sigils, backtracking to the last start sigil which precedes it,
wrapping that text, and then reprocessing the entire set of nodes until
no more end sigils with matching start sigils are found.
This commit implements that behaviour, which is of course a fair bit
more complicated but should improve the results. It also changes
specifying characters by Unicode code·point to use curly braces rather
than angle brackets, as the latter conflicted with the angle brackets
used in links. Specifying by Unicode code·point still isn’t supported
in links, but the behaviour should be less surprising.
The old Unicode code·point behaviour was probably broken also, in the
case where the code·point was not the first character in a paragraph,
but it is fixed now.