Extend and make glob pattern configurable

[skovhus]

Context

Issue: #47

This builds on the work done a year ago (#113). I ended up re-implementing it to get back into the problem space. Thanks to @przepompownia for re-booting this (#156).

This PR

Here we add support for configuring the glob pattern used when pre-analyzing all bash files (this is done to support jump to definition). Besides making it configurable we extend it from **/*.sh to **/*@(.sh|.inc|.bash|.command).

I recommend reviewing commit by commit.

Next steps

This will introduce a few false positives as we do not parse the shebang or mime types (yet), as I tried to document in the fixture folder.

Next step might be to implement the shebang part of #113 or mime types (#47 (comment)) and consider the ideas introduced in #156

But the challenge is that not everyone use shebangs in their bash files... so I guess we shouldn’t make that a filter?

If we can get the included files out of tree sitter, then we could potentially parse those files.

[mads-hartmann]

This is great, thank you @skovhus 🎉

[skovhus]

A few stats to guide upcoming work (shebang/mimetype handling):

Globbing files inside our node_modules (just to have a bunch of files):

1. **/*@(.sh|.inc|.bash|.command) and **/*.sh resolves in 3 seconds (with 3 files)
2. **/* resolves in 5 seconds with 14K files

Reading the file content (sync) of the 14K files takes around 6 seconds. This would need to be done if we analyze the shebang of all files.