feat(code): incremental AST parsing cache

[cniska]

What would you like?

Reuse the parsed AST across sequential edits on the same file instead of re-parsing from scratch each time.

Motivation

Each edit-code call re-parses the entire file. When the model makes multiple edits to the same file in one task, this is redundant. Caching the AST between edits would reduce latency for multi-edit workflows.

Proposed approach

Cache the parsed AST per file path within a task scope. Invalidate when the file content changes after an edit. Measure performance impact before and after.

Scope

• code-ops.ts
• Related test files

[cniska]

Likely premature optimization — AST parsing of single files is fast (milliseconds), so caching across sequential edits within a task would yield marginal savings. The cache invalidation complexity (content hashing, external modification handling, lifecycle scoping) isn't justified without a measured bottleneck. Worth revisiting if profiling ever shows parsing as a hot path.