Optimize docker image

The docker image that's being published for this repository is severely fragmented. As a result, it takes a much longer time to download than it should and consumes a lot of extra disk space. I count nearly 30 layers, which occupies close to 2GB. It really should be just one layer. Please optimize the image by consolidating the RUN commands or by using the `--squash` flag when building.

Also, what's going on with the versions? Here are the versions I see on [DockerHub](https://hub.docker.com/r/algolia/docsearch-scraper/tags):

```
v0.9
v1.1
v1.1.0
v1.2.0
v1.3
v1.3.1
```

Can we stick to a consistent pattern? And how was v0.9 published after v1.2.0?

Even more concerning, v1.3.1 is absolutely not a patch release. It changed the version of Python from 2 to 3 and added 200MB to the size of the image. Can we follow semantic versioning?

I expect better software practices from Algolia than this. Let's work towards that.

Start by inspecting the Docker build configuration and the published DockerHub tags for algolia/docsearch-scraper. Compare the image layers, size, Python version, and tag history; done means a smaller consolidated image and a consistent versioning scheme that reflects the changes.

Written by the indexing model from the issue text.