docs(website): improve search

[etiennebacher]

Close #905

It seems that the index.json used for search cannot take into account pieces of words, e.g sum_hor in pl_sum_horizontal, which explains why the search results were so bad.

This PR adds custom separators when building the index.json for search, meaning that pl_ and other prefixes are no longer taken into account when looking for function names. Note that [\\s\\-]+ is the default, so only what comes after is customized.

Here are some search suggestions before and after this change.

sum_horizontal

Old:

New:

strptime

Old:

New:

sink

Old:

New:

[eitsupi]

I understand. But I don't want to close #686 because this is not a fundamental solution.
Do we need to also add $ as a separator?

[etiennebacher]

I don't want to close #686 because this is not a fundamental solution.

We can close #686 later, I've updated the initial post of this PR.

Do we need to also add $ as a separator?

No because the search index is built on the Markdown files, which still have the _ as separator, e.g DataFrame_clone(). We only replace those by $ in the post-processing script, which is the last thing to run.

[eitsupi]

No because the search index is built on the Markdown files, which still have the _ as separator, e.g DataFrame_clone(). We only replace those by $ in the post-processing script, which is the last thing to run.

I understand that. My point was that since $ is used in places such as examples, shouldn't that also be registered as a separator?

[etiennebacher]

My point was that since $ is used in places such as examples, shouldn't that also be registered as a separator?

I added it, it allows a bit more matches indeed