Keep attributes in their container, fix final newline and reference URL continuation

[dereuromark]

Three smaller differences from djot.js 9790d4d, one commit each.

Block attributes leaked out of their container

::: d
{#x}
:::
tail

djot.js: <p>tail</p>. djot-php: <p id="x">tail</p>. Also - a, blank, {#x}, blank, tail: the list took the blank line, so the reset a blank line does never ran. The syntax description: "put the attributes on the line immediately before the block".

Pending attributes are now dropped when a div closes and when a block's parse ends on a blank line. > {#x} followed by tail still puts the id on tail, as djot.js does.

Unclosed code fence at the end got an extra newline

A fence line alone rendered as <pre><code>\n</code></pre>, and a fence line followed by x as x\n\n. djot.js gives <pre><code></code></pre> and x\n. The cause was the input split: the newline that ends the last line left an empty string behind, which block parsing read as one more line. The line list now stops at the last terminated line.

Reference URL continuation

[r]: /u
  a b

[r][]

djot.js: <p>a b</p> and a link to /u. djot-php linked to /ua b, and > tail gave /u> tail. The syntax description: "None of the chunks of the URL may contain internal whitespace." A continuation line now holds a single chunk.

It also has to be indented past the definition's own [: [r]: /u followed by tail makes tail a paragraph. The description shows only a continuation under a column-0 definition and does not say what the indentation is measured from, so this follows djot.js. A chunk with trailing spaces still continues the URL; djot.js makes it a paragraph, which the description does not support either way.

Measured against djot.js

On the 4853 generated block-family inputs, djot-php differed from djot.js on 1077 before (after #312) and 1041 after: 1046, 1046 and 1041 after each commit. 36 cases moved to matching and none the other way. The fence commit moves no case on its own, because every generated input it touches also differs in another way; the differential's fence family drops from 38 minimized cases to none.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.08%. Comparing base (b038fb5) to head (58a4359).

Additional details and impacted files

@@            Coverage Diff            @@
##             master     #314   +/-   ##
=========================================
  Coverage     93.07%   93.08%           
- Complexity     4732     4737    +5     
=========================================
  Files           118      118           
  Lines         12636    12645    +9     
=========================================
+ Hits          11761    11770    +9     
  Misses          875      875           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.