This will fixes #787

To understand what does not work currently, here is a index.Rmd to reproduce

---
title: "My book"
site: bookdown::bookdown_site
output:
  bookdown::gitbook: 
    split_by: "section"
---

# Cross-references {-}

| Target | Label |
| ------ | ----- |
| Chapter 1 | \@ref(chapter-1) |
| Section 1.1 | \@ref(s:1:1) |

# Chapter 1

## Section 1.1 {#s:1:1}

```{r iris-plot, fig.cap="A plot"}
plot(iris)
```

See figure \@ref(fig:iris-plot)

When using split_by: sections , a h2 following a h1 with no content in between will broke the section reference, and will create an incorrect number in figure reference.
You'll get a

Warning message:
The label(s) chapter-1 not found

This is because bookdown processing will add a closing div tag, to support special bookdown feature, when split_by is section or section+number

When the second header follows directly the first header (i.e x[i] would be this h1 header), it will add this closing \n</div> to the end of the h1 header. This will break parse_section_labels() that won't find the id because the regex does not take into account this addition.

This PR fixes this by detecting the extra div added when split by sections.
Also, I added some tests to check that it will continue to work in the future.