Add troubleshooting page

[penelopeysm]

See #583 for motivation. The idea is to have a dedicated page to the error messages that people receive and provide some guidance. It starts with only one error, but we can add to it as we go along.

Preview here: https://turinglang.org/docs/pr-previews/603/usage/troubleshooting/

I do suppose it's not mandatory to come up with error codes like T0001 (in this case, I chose T because the error comes from Turing.jl; something in DynamicPPL would be D0001, for example).
However, I think that error codes are more stable and easier to search for than error messages (which may be changed, and may vary depending on interpolated variables).
The intent would be that after adding a code here, we can edit the message in the Turing codebase to point to this link

[github-actions]

Preview the changes: https://turinglang.org/docs/pr-previews/603
Please avoid using the search feature and navigation bar in PR previews!

[penelopeysm]

I'll fix the version consistency thing after this PR, since that's just an annoying operation that will require rebuilding the entire docs and takes 2 hours.
Done.

[mhauru]

I think this is a great idea. Might be worth bringing this to @yebai's attention as well, since it's an entirely new aspect of documentation. Some ideas for errors to write troubleshooting bits for:

• The error you get when you forget to do x = Vector{T} and just do x = Vector{Float64} and AD falls over.
• I thought I had a second one in mind but I don't.

[penelopeysm]

Did you have thoughts on this @yebai?

[yebai]

I like the idea of a new doc page for troubleshooting, but the error code will not help much. It reminds me of some weird Windows OS error codes, which are generally not very helpful.

Google is very good nowadays, so searching for an error message is generally not that difficult.

[penelopeysm]

Okay, I don't mind skipping the numeric code -- I'd like to put a direct link to this in the error message, so that should be enough too.