Not only is this tidier, but it avoids making these iterators
Rayon-specific.

This offers a big performance bump for large sites:

  small-kb (100 pages)
    before: 530ms
     after: 500ms

  medium-kb (1000 pages)
    before: 2.9s
     after: 1.5s

  huge-kb (10,000 pages)
    before: 150s
     after:   9.1s

Fix #383

Closes #388

Closes #417

Closes #403

Fix #412

As mentioned in #381, crates.io 404's any request without an Accept:
text/html header.  It 200's any request with one, but at least
false-successes don't prevent checking any other links.

This also makes it easier to add a custom User-Agent if desired.

rustfmt and fix a clippy nit (unnecessary return) while I'm here.

As mentioned in #376

Push summary handling into Markdown parsing, identifying the presence
of one by giving its length in the rendered markup.

Hopefully a better fix for #376.

Also fix tests

Close #367, #372

This reverts commit c7156a84f0.

Replace `populate_next_and_previous_page` with `populate_siblings`

This commit removes the option to sort by order and also removes
`page.next` and `page.previous` variables.  Instead, pages can be sorted
by two methods `date` and `weight`.  The Tera `reverse` filter will
reverse either of those sorts, so the old `order` behavior can be
achieved by using the `reverse` filter with `weight`.

In place of the `previous`/`next` variables, this commit adds the
`page.earlier`/`page.later` variables (which are set when the page is
sorted by date) and the `page.heavier`/`page.lighter` variables (which
are set when the page is sorted by weight).  These variables have the
advantage of not having confusing semantics when the `reverse` filter is
used.

Close #115

Since we are using a builtin template, we can be assured to have the
target file in place. Middleware renamed since it is bound directly to
the concept of handling 404s.