Entity-level code map via sem-core (#315)

[rs545837]

Implements the entity-level code map from #315. Thanks for the enthusiastic green light @ODAncona, excited to land this.

What it does

With --entity-map, code2prompt extracts each file's structural entities (functions, classes, methods) with line ranges and signatures via sem-core, and renders a compact outline before the file contents. So a prompt can lead with the shape of the codebase instead of making the model infer it from full files.

Example output:

Code Map:

`math.py`:
  - class Calculator `class Calculator:` (lines 1-7)
  - function add `def add(self, a, b):` (lines 3-4)
  - function main `def main():` (lines 9-11)

Design decisions (the ones flagged in the issue)

• Placement: both. Per-file via FileEntry.entities, plus a top-level code_map aggregate, since the data is useful both ways and costs nothing to expose twice.
• Format: structured data is the source of truth (name, kind, line range, signature, parent); a default Handlebars partial renders the indented tree, so templates that want a signature-only or custom view can build it.
• Default + flag: off by default, opt in with --entity-map / Code2PromptConfig.entity_map.
• Build cost: sem-core carries tree-sitter grammars for ~28 languages, so it is an optional dependency behind the entity-map Cargo feature. Standard builds are unaffected.
• Privacy: sem-core is the offline library and emits no telemetry (that lives only in the sem CLI binary, not the crate), so code2prompt stays fully air-gapped. Quick proof: grep -r reqwest in sem-core returns nothing.

Try it

cargo run -p code2prompt --features entity-map -- /path/to/repo --entity-map

Tests

Unit tests for Rust and Python extraction (feature-gated); full suite passes with and without the feature; verified end to end on a mixed sample.

Scope

This is the code map. The natural follow-up is function-level context forging (pull a function plus its real callers/callees across files, token-budgeted, via sem's dependency graph), which is the bigger token-savings win. Happy to do that as a second PR once this lands.

[ODAncona]

LGTM,

I'll take over with the chores =>

• Update doc
• Move unit tests in /test
• Remove some comments

I'll try it out and come back to you for next steps