Improve grep perf planning, benchmarks, and ordered emission

[niamster]

Summary

• add an ADR for the grep performance plan and benchmark prerequisites
• add deterministic Criterion benches for pattern matching, traversal, sparse search, and end-to-end walk plus search
• refactor the walker to collect files globally, sort them lexicographically, and emit results in ordered sequence while still searching files in parallel

Benchmarks

Criterion comparison between the benchmark baseline commit c067e96 and the refactor commit c79f61b:

• patterns_double_star_match: 15.389 ns -> 15.671 ns
• walk_ignore_and_filter_only: 917.15 us -> 940.71 us
• search_sorted_sparse_file_list: 9.1030 ms -> 9.4201 ms
• walk_and_search_dense_tree: 15.078 ms -> 14.999 ms

External hyperfine run on /System/Library from the current branch release binary:

• PATH tgrep: 18.273 s
• local ./target/release/tgrep: 12.939 s

Testing

• cargo test
• cargo bench --no-run
• cargo bench --bench patterns -- --noplot