Camelot returns tables that contain no text (Where text should be detectable)

I'm trying to extract data from some ~900 certificates. These certificates have an identical visual structure, but are published by different parties. For the majority of files the extraction works. However, for several dozen files, the table-structure returned by Camelot contains only empty strings.

Plotting `grid` and `text` shows content is detected (e.g. Table 7 in [DS_3663.pdf](https://github.com/camelot-dev/camelot/files/10046174/DS_3663.pdf)):


I'm using this command to read the pdf and create the tables:
`>>> tables=camelot.read_pdf('pdfs/DS_3663.pdf', pages='1-end', line_scale=110, shift_text=[''])`

e.g. Table 7 contains this data:
`>>> tables[7].data`
`[['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', ''], ['', '', '', '', '', '', '', '', '', '', '', '', '', '']]
`

Here are a few more example pdfs where the extraction fails in an identical manner:
[DS_885.pdf](https://github.com/camelot-dev/camelot/files/10046200/DS_885.pdf)
[DS_2481.pdf](https://github.com/camelot-dev/camelot/files/10046214/DS_2481.pdf)
[DS_2083.pdf](https://github.com/camelot-dev/camelot/files/10046204/DS_2083.pdf)

Parsing all of these files with `pdf2txt.py` successfully extracts text, so I assume it should be possible to get a result with Camelot as well.

**Environment**

- OS: Ubuntu 22.04.1 LTS
- Python version: 3.10.6
- Numpy version: 1.23.4
- OpenCV version: 4.6.0.66
- Ghostscript version: 9.55.0
- Camelot version: 0.9.0

I've tried debugging this, but had difficulties understanding the intricate code in the bbox-sections. From what I've figured out, it appears to me that Camelot is unable to marry `horizontal_text` (Which contain the relevant text) with the line-grid.

Reproduce the empty-table result with camelot.read_pdf on DS_3663.pdf, especially Table 7, and compare it with DS_885.pdf, DS_2481.pdf, or DS_2083.pdf. Start by tracing how horizontal_text is associated with the line-grid in the bbox-sections code. Done means these PDFs return their detected text in tables[7].data rather than rows of empty strings.

Written by the indexing model from the issue text.