[rupert648]

Summary

• Moves MACHINERY_NEEDLES / is_machinery into the core culpert crate (pprof.rs) so frame filtering happens at encode time, not just in CLI display
• backtrace::trace, culpert::sampler, alloc::raw_vec:: and similar internal frames are now absent from the emitted .pb.gz — stock go tool pprof / pprof -text shows real user code at the leaf without any scaffolding
• Samples whose entire stack is machinery frames are preserved (bytes intact) but emitted with an empty location_id list rather than dropped
• CLI's format_leaf_callsite and its own copy of MACHINERY_NEEDLES are kept as-is for backward compatibility with existing profiles on disk

Test plan

• pprof::tests::machinery_frames_stripped_from_encoder — verifies machinery frame names don't appear in any encoded Function; verifies user frames survive
• pprof::tests::all_machinery_stack_emits_empty_location_list — verifies all-machinery stack emits the sample with correct byte count but zero locations
• All existing unit + integration tests pass (cargo test --workspace)
• cargo clippy --all-targets --all-features clean
• cargo fmt --all clean

🤖 Generated with Claude Code

[github-actions]

culpert: allocation diff

• before: /tmp/culpert-baseline.pb.gz — total 12.02 MB (estimated)
• after: /tmp/example-macros.pb.gz — total 12.09 MB (estimated)
• net Δ: +67.97 KB (+0.55%)
• sample rate: 4.00 KB/alloc
• filter: show changes ≥ 1.00 MB AND ≥ 10.00%

Regressions

None.

Improvements

None.

_{2 span(s) suppressed by thresholds.}