fix: handle empty LSP glob patterns

[djdv]

fixes: #672
I have a screencast of the issue + this patch.

The LSP spec for glob patterns defines the syntax for "group sub patterns" but doesn't mention anything about empty patterns.

Somehow (not investigated) we end up with an empty s parameter being passed to our glob function, so I assume that since such a pattern isn't explicitly forbidden, that language servers (TSserver, marksman, et al.) emit them?
(Maybe this is worth checking up front if s is empty, but if #filtered == 0 then catches this. Might be even better to not call this function at all in some higher scopes but I'm not sure.)

Imagine patterns such as: **/*.{md,}, **/*.{}, **/*.{foo,,bar}.
These patterns get rejected by the parser for vim.glob.to_lpeg(): Invalid glob: **/*.{}.

---

This change replaces glob so that patterns are always sanitized and conditionally sorted before being passed to to_lpeg().
So that *.{} is treated like *, *.{md,} -> *.{md}, *.{foo,,bar} -> *.{foo,bar}.
I believe these are all permissible in the current spec, so we account for them.

---

There may be a better way to solve this, I'm not sure.
This works for me in my use case (NVIM v0.12.0-dev-1908+g47aef025a7 on Windows) but could use some testing from others.
Older vim users, people reporting this issue on other platforms + LSP servers and scenarios, etc.
Hopefully this approach doesn't break any existing workflows.

[stevearc]

Seems fine to me. I made one tweak to make the pattern non-greedy (a bug with my original implementation). Thanks for the PR!