Problem: fugitive shows combined diffs (@@@ headers, 2-char prefixes)
for unmerged (UU) files. The parser and highlight pipeline assumed
unified diff format (@@, 1-char prefix), causing broken prefix
concealment, missing background colors on ` +`/`+ ` lines, and no
treesitter highlights due to garbage prefix chars in code arrays.
Solution: detect prefix width from the number of leading @ signs in
hunk headers. Propagate prefix_width through the parser (new field on
diffs.Hunk) and highlight pipeline (prefix stripping, col_offset,
concealment, line classification). Add U to filename pattern for
unmerged file detection. Skip intra-line diffing for combined diffs
since the 2-char prefix semantics don't produce meaningful change
groups.