Incremental re-parse, snapshot structural sharing, and structural diff

[konard]

Context

Issue #47's title demands "the richest features set". The competitor survey
(competitors-code-tooling.md) ranks
incremental re-parse with subtree reuse (tree-sitter) and red/green persistent
trees with structure sharing (Roslyn, rowan/Biome) as the top capabilities
meta-language lacks; difftastic adds structural snapshot diffing. Today every
edit re-parses from scratch. See
solution-plans.md S-13.

Scope

• Expose tree-sitter's native incremental parsing (InputEdit + previous
  tree) through src/tree_sitter_adapter.rs:
  LinkNetwork::apply_edit(range, new_text) re-parses only affected regions;
  link ids outside the edited span stay stable.
• Extend NetworkSnapshot so an edited fork shares unchanged links with its
  parent (rowan/cstree precedent), with a test measuring sharing.
• Structural diff between two snapshots returning changed/added/removed link
  sets (difftastic precedent) - this also gives transforms a cheap dry-run
  preview.

Acceptance criteria

• apply_edit produces a network equal to a from-scratch parse of the
  edited text (byte-exact reconstruction) while preserving ids outside
  the edit.
• Snapshot fork sharing is asserted (unchanged links are not duplicated).
• Structural diff is tested on an edit fixture and exposed publicly.
• Changelog fragment added (bump: minor).

References

• Requirement: issue title ("richest features set"); requirements.md R-2 context
• Solution: solution-plans.md S-13
• Part of We need to carefully compare all competitors in all scopes with out project, and make sure we have the richest features set #47; work lands on branch issue-47-76af108c0f24 (PR Finish issue #47 parity feature set #48).

---

Filed from docs/case-studies/issue-47/proposed-issues/13-incremental-reparse-structural-sharing.md. Part of the implementation plan for #47.