docs: performance section with measured benchmarks and jq comparison

[alvinreal]

Summary

Upgrades the README performance section from generic claims to reproducible, measured benchmarks and real head-to-head comparisons.

What was benchmarked

All tests use the same 10,000-record dataset on the same machine (macOS arm64), with warmup + repeated runs via hyperfine.

1) JSON → YAML conversion

• morph: 23.7 ms
• yq: 713.2 ms
• Result: morph 30.03x faster

2) JSON transform (rename + filter)

Task: rename .name -> .username + where .age > 30

• morph: 18.3 ms
• jq: 40.0 ms
• yq: 101.5 ms
• Result: morph 2.19x faster than jq, 5.55x faster than yq

3) CSV → JSON conversion

• morph: 11.7 ms
• miller: 17.2 ms
• Result: morph 1.47x faster

Also included

• Updated Criterion throughput table (parsing/conversion/mapping)
• Exact benchmark commands in README for reproducibility

Verify locally

cargo bench --bench benchmarks

And run the hyperfine commands listed in the README head-to-head section.