MD038: judge the multi-line code spans whose literal alone proves padding

#402 stopped MD038 from judging a code span whose delimiters sit on different
lines. Nothing recorded about such a span describes what was written against its
delimiters: `literal` has already turned the line endings into spaces and taken
the padding off, a width cannot be had from columns that index two different
lines, and reading the lines back fails four ways that #402 documents. Skipping
is right for those four shapes, but it also gives up on a subset that `literal`
alone can still prove.

## The subset

A line ending contributes exactly one space to `literal`. So whitespace deeper
than one byte at either end cannot have come from a line ending, and has to be
in the source.

```markdown
`a
b `
```

comrak gives `literal = "a b "`. Two trailing spaces, one of which is at most
the line ending, so at least one was written. Same the other way:

```markdown
` a
b`
```

gives `literal = " a b"`. mado reported both of these before #402 and passes
them silently now.

## The guard has to be the two-space one

A `trim()`-style test is **not** safe here:

```markdown
> `text
>`
```

gives `literal = "text "`, and that single trailing space is the line ending in
its entirety — there is no padding in the source at all. Reporting it would be
exactly the kind of violation an author cannot act on that #402 was about.

"Two or more spaces at an end" is the version that holds, because the line
ending can only ever account for one of them.

## What is left open

- Whether one space of padding either side of a line ending — the case this
guard cannot separate from the line ending itself — is worth any further
effort, or whether the two-space subset is where this should stop.
- Tabs. `literal` keeps a tab as a tab, and a line ending never becomes one, so
a leading or trailing tab is provably from the source even on its own. Whether
MD038 should treat a tab as padding is a separate question from whether it can
see it.
- Where the violation should be reported. The columns of a multi-line span are
not reliable (#403 is a related shift), and the two ends of one belong to
different lines.