chore: cover every reachable branch and drop the unreachable ones

[hownowstephen]

Coverage goes from 97.7% to 99.8%. The one statement left is the error return from html.Parse. It only fails when its reader does, and a strings.Reader never fails, but Go has no way to exclude a line from coverage and discarding the error would be worse.

Removed as unreachable:

• the n == nil guard in doConvert. No caller passes nil.
• the missing-marker breaks in extractPre and restorePre. Both now use strings.Cut, since the markers are always written in pairs and one placeholder is written per block. Merge fix: keep control characters in content from acting as markers #81 first, because that is what stops content from injecting unpaired markers.
• target >= total in WordWrap. The loop only runs while endRune + lineLength < total, so it never held. The remaining clamp is a single max.

New tests for branches that were reachable but untested:

• <br> inside <pre>
• an empty anchor with footnote links
• an <img> nested inside an element within a link
• a style with a trailing ;
• WordWrap past a run of spaces longer than the line
• a ) whose only earlier break is the one being moved

The new WordWrap was compared against the one on main over 300,000 random inputs covering brackets, runs of spaces, multi-byte runes and widths from -1 to 10, and output was identical. Output is byte-for-byte identical to main across the corpus under every link style. FuzzFixSpacing passes.