highlight() is inconsistent when code is illegal syntax

[seaoak]

Generally the result of highlight() contains lauguage name as HTML class name.

If code is illegal syntax for the specified language,
the HTML class name is added or not (inconsistent).

option lang	illegal syntax?	option hljs	class name of outer <figure>	class name of inner <code>
java	no	false	highlight java	N/A (none)
java	no	true	highlight java	hljs java
java	yes	false	highlight	N/A (none)
java	yes	true	highlight	hljs java
none	-	false	highlight plain	N/A (none)
none	-	true	highlight plain	hljs plain

I think we have three options:

1. always add laugage name even if illegal syntax.
2. never add language name if illegal syntax.
3. add the language name plain if illegal syntax.

I think Option 1 might be best because users will intend to be.

Please advice!

---

I think this inconsistency is caused by undocumented behavior of highlight() of highlightjs package.

The API document of highlightjs package is:
https://highlightjs.readthedocs.io/en/latest/api.html#highlight-languagename-code-ignore-illegals-continuation

According to above API document, highlight() throws an exception in case of detecting illegal syntax for the language.
But actually not throw any exception (regardless ignore_illegals parameter).
Instead, illegal property of the return value is set to true and
language property of the return value does not exist.

Current code of lib/highlight.js considers the exception halfway.
And the lack of lanaguage property is not expected.

Anyway the code should be modified to keep consistency.

p.s. This issue is derived from hexojs/hexo#3984

[stevenjoezhang]

@seaoak I agree that option 1 might be the best choice. Can you help test whether this pull request meets your expectations? #246