The markdown renderer passes its output through an HTML pretty printer.
While this is good in most cases, it makes a mess of elements that expect
their content to be left untouched.
The pretty printer already ignores `pre` tags (and other built-ins) by
default. This fix allows us to specify other tags that should be left
alone.
Further it actually specifies this option for `code-example` and `code-pane`
tags, which expect to contain preformatted content.
This processor will eventually replace the `{@example}` inline tags
because it provides a cleaner approach that also supports tabbed examples
straight out of the box.
The idea is that authors will simply add a `path` and (optionally) a `region`
attribute to `<code-example>` or `<code-pane>` elements in their docs.
This indicates to dgeni that the relevant example needs to be injected
into the content of this element.
For example, assume that there is an example file `toh-pt1/index.hml` with
a region called `title`, which looks like:
```
<h1>Tour of Heroes</h1>
```
Then the document author could get this to appear in the docs as a
standalone example:
```
<code-example path="toh-pt1" region="title"></code-example>
```
Or as part of a tabbed group:
```
<code-tabs>
<code-pane path="toh-pt1" region="title"></code-pane>
</code-tabs>
```
If no `path` attribute is provided then the element is ignored, which
enables authors to provide inline code instead:
```
<code-example>
Some <html> escaped code
</code-example>
```
Also all attributes other than `path` and `region` are ignored and passed
through to the final rendered output allowing the author to provide
styling hints:
```
<code-example path="toh-pt1" region="title" linenums"15" class="important">
</code-example>
```
Previously, all URLs were rewritten to `index.html` in order to support
deep-linking. This works when navigating to URLs that correspond to existing
resources. E.g. navigating to `/tutorial` returns `index.html` and then the
`DocViewer` takes over and requests `tutorial.json`.
Navigating to a non-existent URL (e.g. `/foo`), will return `index.html`, which
in turn requests (the non-existent) `foo.json` and throws an error when trying
to parse the returned `index.html` as JSON.
This commit fixes it by only rewriting URLs that do not request a file (i.e. do
not include a `.` in the last path segment).
Fixes#15398
