nx_eqsat_dsl_bench.nx

nx_eqsat_dsl_bench.nx -- RULE-HEAVY race: the sovereign NishiLang DSL e-matcher vs egg 0.11 (race_egg/examples/nx_dsl_bench.rs), SAME expression, SAME 4 rules, engine-only (build->saturate->recompute_best->extract->read-root), honest us/op. The DSL e-matcher (nx_eqsat_apply_dsl_table over a DATA table) drives saturation -- the same general matcher proven byte-equivalent to the inline rules and proven to apply the new (xor x x)==0 data rule. Here it runs the 4-rule subset that egg races (add_zero, sub_self, mul_one, mul_pow2) so the e-matching work is 1:1. Rule-heavy expression (touches 4 rules in ONE graph): (+ (* (- y y) 1) (* x 8)) (- y y) -> 0 (sub_self) (* 0 1) -> 0 (mul_one) (+ 0 (* x 8)) -> (* x 8) (add_zero) (* x 8) -> (<< x 3) (mul_pow2) best extracted = (<< x 3) FAIL LOUD: if the extracted root is not SHL the harness sys_exit(nonzero) BEFORE timing -- a fast wrong answer is not a win. Same honesty discipline as the existing nx_eqsat_race_bench.nx (egg's exact pipeline, no proxy, no proof machinery in the timed loop -- egg has none). Output (one line): NX_DSL_BENCH iters=100000 total_us=<t> us_per_op_ns=<u>