Optimize unifier pair handling

[MesTTo]

Reduces work in the expression unifier without changing unification semantics.

The branch drops identical dereferenced pairs before descending them and uses HashSet::insert as the seen-pair guard instead of hashing once for contains and again for insert.

Validation:

• cargo +nightly-2026-06-01 test -p mork-expr with RUSTFLAGS='-C target-cpu=native'
• cargo +nightly-2026-06-01 test -p mork --lib with RUSTFLAGS='-C target-cpu=native'
• jscpd on expr/src/lib.rs
• git diff --check

[MesTTo]

Closing per the kernel gate on constant-factor speed-ups (same basis as #126/#127): this reduces unifier work without changing semantics (dropping identical dereferenced pairs, hashing the seen-pair guard once instead of twice), a constant-factor optimization rather than an asymptotic one. The asymptotic version would be replacing the (ExprEnv,ExprEnv) hash-set guard with a union-find trail; parking this micro-opt.