[stacked] Make parser, highlighting, and merge handling rail-style aware

[barrettruth]

Part of #358.

Goal

Make rail consumers style-aware so parsing, highlighting, and merge/conflict helpers work correctly with both dual and single rails.

Scope

• Teach parser.parse_buffer() to read generated rail style and store it on parsed hunks where needed.
• Update highlighting to use style-aware rail number range helpers.
• Ensure hide_prefix overlays never cover single rail line numbers.
• Ensure DiffsRail, DiffsRailNr, DiffsAddRailNr, and DiffsDeleteRailNr apply correctly in single rail buffers.
• Ensure merge/conflict helpers strip single rails correctly before matching or resolving hunks.
• Keep all raw hunk lines and intra-line diff logic rail-free.

Non-goals

• Do not change hunk action semantics.
• Do not change treesitter or intra-line diff algorithms except where column offsets must account for rail style.
• Do not change native split endpoint windows.

Acceptance criteria

• Parsed hunks from single-rail generated buffers match raw unified diff hunks.
• Single rail highlighting is aligned and uses the expected highlight groups.
• Merge/conflict parsing strips single rails and still resolves to working-file conflict regions.
• Existing dual-rail parser/highlight/merge tests continue to pass.

Verification

• Add parser tests using single-rail display lines.
• Add highlight tests for single-rail ranges and prefix concealment.
• Add merge/conflict tests for single-rail generated merge diffs.