perf(meta_generator): re-implement as a helper

[curbengh]

What does it do?

From the benchmark performed by @SukkaW , we can see cheerio—used by meta_generator filter—incurs significant perf cost. The feature was initially introduced by #3129.

Despite effort to improve the situation (#3667), the improvement was only minuscule.

As such, I propose to re-implement it as a helper instead, inspired by open_graph(). This approach doesn't require any dep, so it shouldn't incur perf cost.

However, as mentioned in #3663 (comment), this approach also means meta_generator will instead rely on adoption by the themes. Meaning, the adoption will be back to square one.

How to test

git clone -b metagen-helper https://github.com/curbengh/hexo.git
cd hexo
npm install
npm test

Pull request tasks

• Add test cases for the changes.
• Passed the CI test.

[coveralls]

Coverage decreased (-0.006%) to 97.148% when pulling 8c33a0f on curbengh:metagen-helper into 3176fe0 on hexojs:master.

[curbengh]

I just thought of a less breaking approach.

Edit: #3671

[curbengh]

Superseded by #3671