Don't decode url before encoding it again

[danielberndt]

Decoding before encoding leads to invalid urls. Here's an example:

Let's assume this markdown:

[link](https://www.example.com/home/%25batty)

Note: %25 is the code for the % character

This results in:

var dest1 = decode("https://www.example.com/home/%25batty")
//dest1 = https://www.example.com/home/%batty

var dest2 = encode(dest1)
// dest2 = https://www.example.com/home/%batty 

since %ba is a potential percentage-encoded hex code, mdurl's encode doesn't change it. resulting in an invalid link.

As far as I'm aware, just doing encode(uri) is doing the desired thing. for all kinds of uris

[jgm]

Test failure is

Regression tests  ✓✓✓✓✓✓✓✓✓✓✓
250✘ Example 12
251=== markdown ===============
252[XSS](javascript:alert%28'XSS'%29)
253=== expected ===============
254<p><a␣href="javascript&amp;colon;alert('XSS')">XSS</a></p>
255=== got ====================
256<p><a␣href="javascript&amp;colon;alert%28'XSS'%29">XSS</a></p>

It seems to me that the change in output is unproblematic; do you agree?
If so, we can change the expected output for this test.

[colinodell]

Could we please also add the broken example in this PR (https://www.example.com/home/%25batty) to the regression tests?

[danielberndt]

Alright, I just added a fix to the existing regression test as well as adding a %25 based test

[danielberndt]

I also just removed the decode import, so as a side effect this PR should make this library a little bit smaller and faster :)

(fyi: there seems to be no lint rule warning about unused variables)

[jgm]

Thanks! And thanks for the lint suggestion. I added the rule and found another unused variable.