Change repo cloning, file splitting and general refine, second attempt

[AlessandroAnnini]

Hello @gcapuzzi,
this is a second attempt after PR #2

I changed how the repo is cloned: no more github api and no more need to use a github PAT but just a simple git clone running a bash command in a subprocess.

The files are filtered by a list of allowed extensions called ext_whitelist, this way you can grab md, mdx, and other files all together.

The files that pass the filter are splitted using a specific splitter for markdown files that creates metadata about the titles in the file itself, this is good for the quality and i think the speed of the research in the vector db. But the problem is that it works only with markdown, so, if your ext_whitelist has .js files, those should actually use a different splitter. The splitter selection should be based on the extension and more splitters should be used dynamically.

I took a stab at using Langchain LCEL but I couldn't implement memory the way that I wanted so it is commented.

The rest is just a refine but nothing important.

p.s. keeping the code in a notebook is pretty inconvenient because in this PR you cannot really see what I have changed unless you go to my project and try that, the history of the changes in the file is non-existent too.