Task: ores.diff: intra-value diff spans

Extend ores.diff::domain::diff_entry with intra-value diff spans (character/token ranges within old_value=/=new_value that actually changed) and compute them in ores.diff::engine::compute, per History Diff Architecture. This is the one layer every downstream task depends on: no entity knowledge required, pure string algorithm work, computed once and rendered identically by every frontend.

• diff_span (offset/length) added; diff_entry carries old_spans=/=new_spans.
• engine::compute populates spans for every changed entry: common- prefix/suffix for single-line values, line-by-line LCS plus per-line token diff for multiline values.
• Test coverage (per the redesign task's Notes and the architecture doc): single-character change, whole-word change, common-prefix- only, common-suffix-only, no-overlap, multiline LCS, per-line token diff within a changed multiline block, added field (empty old_value), removed field (empty new_value), identical values (no spans), empty-vs-non-empty string, unicode/multi-byte content (spans never split a code point).
• Existing field-level compute tests stay green. Several previously asserted full diff_entry struct equality, which now also compares the newly-populated span vectors — narrowed those to the specific fields they intended to test (field_name/old_value/new_value) rather than asserting exact spans incidentally.

1. Add diff_span to projects/ores.diff/include/ores.diff/domain/; extend diff_entry with old_spans=/=new_spans.
2. Implement span computation as a private helper inside ores.diff::engine::compute: detect single-line vs multiline by the presence of \n; single-line gets common-prefix/suffix spans; multiline gets a line-level LCS diff, then per-line token spans within changed lines.
3. Write the test list above in projects/ores.diff/tests/engine_compare_tests.cpp (or a new diff_span_tests.cpp if the file grows unwieldy).
4. Verify existing tests still pass unmodified.

Implementation was written without a local build initially (per instruction not to build that session): correctness of the byte-offset arithmetic, including the UTF-8 boundary trimming and the LCS line-alignment, was verified by hand-translating the C++ algorithm into an equivalent Python script and running every test case's expected values through it before writing the C++ tests. Built and tested against linux-clang-debug-make once building was authorised; all 29 test cases (113 assertions) passed on the first run, confirming the Python-verified arithmetic. A CMake reconfigure was required for the new engine_diff_span_tests.cpp file to be picked up by GLOB_RECURSE — the initial build only ran the pre-existing 14 tests.

Manual QA scenarios (scaffolded via compass add test_scenario, run through the QA Validation Runner panel) that verify this task. Link new ones here as they're created; the scenario doc itself links back via its "Verifies task" field.

Added diff_span (projects/ores.diff/include/ores.diff/domain/diff_span.hpp) and extended diff_entry with old_spans=/=new_spans. Implemented span computation in ores.diff::engine::compute (projects/ores.diff/src/engine/compare.cpp): UTF-8-boundary-safe common-prefix/suffix diffing for single-line values, and a line-level LCS diff (positionally pairing same-sized runs of removed/inserted lines, with leftover unpaired lines marked fully changed) plus per-line token diffing for multiline values.

Added projects/ores.diff/tests/engine_diff_span_tests.cpp with 15 tests covering the full acceptance list. Narrowed 6 pre-existing tests in engine_compare_tests.cpp from full diff_entry struct equality to the specific fields they intended to test, since spans are now always populated on changed entries. All 29 tests (113 assertions) pass against linux-clang-debug-make.