Add PathMap benchmark coverage

[MesTTo]

Adds benchmark coverage for graft behavior, node layout, prefix and product cases, streaming cases, write zipper cases, and owned zipper head cases.

The grouped graft benchmark exposed an existing graft_child_maps panic at zipper node boundaries. The branch now also fixes with_node_at_path so it does not ask a trie node for a child using an empty node key when the zipper is already at a node boundary.

Validation:

• cargo check --benches
• cargo test write_zipper_graft_child_maps_zero_child_with_root_value --lib
• cargo test --lib --tests
• cargo bench --bench graft_child_maps_criterion -- graft_child_maps/grouped_keep_unset/contiguous/1 --warm-up-time 0.05 --measurement-time 0.1 --sample-size 10
• jscpd --reporters ai --format rust src/write_zipper.rs

[adamv-symbolica]

Fails to run for me with

Benchmarking graft_child_maps/grouped_keep_unset/contiguous/1: Warming up for 150.00 ms
thread 'main' (9746966) panicked at src/dense_byte_node.rs:637:28:
index out of bounds: the len is 0 but the index is 0
stack backtrace:
   0: __rustc::rust_begin_unwind
   1: core::panicking::panic_fmt
   2: core::panicking::panic_bounds_check
   3: <pathmap::trie_node::tagged_node_ref::TaggedNodeRefMut<u32, ()>>::node_get_child_mut
   4: <pathmap::write_zipper::WriteZipperCore<u32, ()>>::graft_child_maps::<alloc::vec::Vec<pathmap::trie_map::PathMap<u32>>>
   5: graft_child_maps_criterion::grouped_graft
   6: <criterion::bencher::Bencher>::iter_batched::<(pathmap::trie_map::PathMap<u32>, alloc::vec::Vec<pathmap::trie_map::PathMap<u32>>, pathmap::utils::ByteMask), usize, graft_child_maps_criterion::grouped_child_map_grafting::{closure#2}::{closure#0}, graft_child_maps_criterion::grouped_child_map_grafting::{closure#2}::{closure#1}>
   7: <criterion::routine::Function<criterion::measurement::WallTime, graft_child_maps_criterion::grouped_child_map_grafting::{closure#2}, graft_child_maps_criterion::GraftFixture> as criterion::routine::Routine<criterion::measurement::WallTime, graft_child_maps_criterion::GraftFixture>>::warm_up
   8: <criterion::routine::Function<criterion::measurement::WallTime, graft_child_maps_criterion::grouped_child_map_grafting::{closure#2}, graft_child_maps_criterion::GraftFixture> as criterion::routine::Routine<criterion::measurement::WallTime, graft_child_maps_criterion::GraftFixture>>::sample
   9: criterion::analysis::common::<criterion::measurement::WallTime, graft_child_maps_criterion::GraftFixture>
  10: <criterion::benchmark_group::BenchmarkGroup<criterion::measurement::WallTime>>::bench_with_input::<criterion::benchmark_group::BenchmarkId, graft_child_maps_criterion::grouped_child_map_grafting::{closure#2}, graft_child_maps_criterion::GraftFixture>
  11: graft_child_maps_criterion::main

[MesTTo]

I pushed ad91856 for this.

The benchmark exposed a real bug in with_node_at_path. When the zipper is already at a node boundary, self.key.node_key() is empty, but the helper still called node_get_child_mut(key). DenseByteNode::node_get_child_mut expects a non-empty key. In debug builds that trips the assertion; in the bench profile it reaches the key[0] bounds check and panics with the out-of-bounds error you saw.

The fix skips the child lookup when the current node key is empty and uses the existing direct mutation path instead. I added write_zipper_graft_child_maps_zero_child_with_root_value to cover the zero child case with a root value and nested child data.

Checked locally:

• cargo test write_zipper_graft_child_maps_zero_child_with_root_value --lib
• cargo test --lib --tests
• cargo bench --bench graft_child_maps_criterion -- graft_child_maps/grouped_keep_unset/contiguous/1 --warm-up-time 0.05 --measurement-time 0.1 --sample-size 10

[MesTTo]

confirmed gone. ran the exact bench on the parent commit and it still panics at dense_byte_node.rs:637, then on ad91856 it runs clean (grouped_keep_unset/contiguous/1 ~5.6µs).

the panic is just an empty key reaching node_get_child_mut. in release it's the key[0] index that blows up (what you saw), in debug the debug_assert!(key.len() > 0) in that same fn fires instead. nothing exercised a grouped graft sitting exactly on a node boundary before, so it slipped past the existing tests... write_zipper_graft_child_maps_zero_child_with_root_value locks that shape now.