feat(extraction): add Extractor::extract_file_with_source

[nerdalytics]

Description

Extractor::extract_file reads its own bytes with std::fs::read, so a caller that already holds the content cannot use it. I hit this in #78, where I extract each commit out of a bare clone. The content comes from git show <ref>:<path>, and a bare clone has no worktree to read it from.

Type of Change

• New feature (non-breaking change which adds functionality)

Changes Made

Extractor. extract_file now reads the bytes and passes them to a new extract_file_with_source, which contains the original body unchanged. Disk stays the default, extract_file keeps its signature, and the change adds no dependency.

Naming. I followed GraphBuilder::ensure_file_node_with_source (graph_builder.rs:228), which already splits the same way in this crate.

Testing

extract_file_with_source_needs_no_file_on_disk extracts a Rust symbol from a path under a TempDir that was never written, so it fails if the method ever reads from disk again. cargo test -p rgctl-extraction --lib reports 47 passed, 0 failed.

Two pre-existing problems show up in the wider gates, and neither comes from this change. cargo test -p rgctl-extraction --test markdown_context_gql gives 3 passed and 8 failed, and cargo clippy --lib --bins -- -D warnings gives 7 collapsible_if errors in rgctl-plugin-api under clippy 1.98.0. I stashed my diff and reran both against main at 68e3dcc, which gave the same 8 failures and the same 7 errors.

• Unit tests added/updated
• Integration tests added/updated
• Tested with cargo test
• Manual testing performed

Test environment:

• OS: macOS 26.6.2 (Darwin 25.6.0)
• Rust version: rustc 1.98.1 (48a229cea 2026-09-01)

Checklist

• My code follows the style guidelines (ran cargo fmt)
• I have performed a self-review of my code
• I have commented my code where necessary
• I have added tests that prove my fix/feature works
• All new and existing tests pass (cargo test). Eight tests in this crate fail on main too. See the Testing section.
• No clippy warnings. cargo clippy -p rgctl-extraction --all-targets reports nothing in extractor.rs.
• Documentation updated (if applicable). The new method carries a doc comment, and no guide describes Extractor directly.
• CONTRIBUTING.md guidelines followed
• All commits are signed and include DCO sign-off

Performance Impact

• No measurable performance regression

extract_file moves a Vec<u8> it already owned, and does the same single read as before.

Breaking Changes

None. extract_file keeps its signature and its behavior.

Additional Notes

This does not make extraction content-agnostic on its own. FileDiscoverer::discover (rgctl-extraction/src/discovery/mod.rs:80) reaches the filesystem in three places. It walks the tree with WalkBuilder, stats each entry in is_too_large (:153), and opens each file in is_binary (:158). IncrementalUpdater::apply_change_set calls it (rgctl-incremental/src/updater.rs:153). After this change, extract_file is the only content read a caller can bypass, and I am not changing the discoverer here.