Refactor the process

[hupili]

Github's MD API will strip some html tags if they think that not suitable. It breaks my TOC and I just fixed in the commit https://github.com/hupili/evermd/commit/cae362f6442f8a508ccf0ce824aa12eef9bcff85

Prior to this one, I already suffered a case that GH strips my css link.

This fix makes me realize that a refactor is needed for the evermd parser. The original organization is not ideal. The right flow is:

• Preprocess with evermd. Preserve everything created by evermd. Only leave a <pre> with our hash value there.
• Pass the preprocessed doc to any MD processor. (configured by the users).
• Postprocessing by putting back those preserved texts.

[hupili]

Note down GH MD API's problem:

• Strip css using <link>
• Strip id=xx attributes from you A tags.