Benchmark refresh: structural CLI path and tagma fixture numbers

[metasmile]

Goal

Refresh the benchmark suite after the CLI structural integration (issue #42, merged) and the Tagma hardware fixtures (issue #47, merged). Add a benchmark for the actual CLI verify path (evaluate_structural, which the CLI now runs by default) and for the new Tagma decoder fixture, then re-measure on merged main and update the documented numbers.

Scope

1. Add evaluate_structural/cva6_full_33M to benches/bench.rs: this is the pipeline ev verify runs today (raw total + StructuralEnum + evaluate_all on the valid subset), currently cited in the README as ~0.2 s but not benched directly.
2. Add Tagma fixture benches: evaluate/tagma_decoder_64K (naive, 65,536 raw) and struct_enum/tagma_decoder_64K (11,172 valid, 17% density).
3. Re-run cargo bench on merged main and update the fixture reference table and per-function comments in benches/bench.rs, plus the README validation table, with the fresh measurements.
4. Fix the stale docs/cva6.qmd reference in the bench header (the Quarto docs were removed).

Acceptance criteria

• cargo bench compiles and runs on merged main.
• The bench header table and README cite measurements taken after the merges.
• The new evaluate_structural bench reflects the CLI verify path, so the README's ~0.2 s claim is reproducible as a benchmark.

References

• issue CLI structural integration + CVA6 full struct_enum benchmark verification #42 (CLI structural integration, merged via PR PR: feat: make structural enumeration the CLI verify default #42 #48)
• issue Tagma hw RTL fixture specs: feasible verification YAML specs and plan #47 (Tagma fixtures and coverage gate, merged via PR PR: feat: tagma hardware fixture specs, decode projector, and coverage gate #47 #49)
• benches/bench.rs fixture reference table