[ubelai]

Summary

Fixes #730

While inclusive range loops (for v in lo..=hi) already hoist both bounds (hyLo and hyHi) before declaring the loop variable v, exclusive range loops (for v in lo..hi) previously emitted:

for v := <lo>; v < <hi>; v++ {

This caused two bugs:

1. Outer-variable shadowing (for n in 0..n): In Go's for header scoping rules, n := 0 in the init statement is in scope for the condition n < n, so for n in 0..n lowered to for n := 0; n < n; n++ (0 iterations instead of reading the outer n).
2. Repeated upper-bound evaluation: Side-effecting or expensive hi expressions (for i in 0..expensive()) were re-evaluated on every iteration instead of once at loop entry.

Changes

1. compiler/crates/hy-emit/src/stmt.rs:
   • Hoist the exclusive range upper bound hi_t into a fresh hyHi temporary (let hi_once = self.next_temp("hyHi"); self.push(&format!("{} := {}", hi_once, hi_t));) before emitting for {} := {}; {} < {}; {}++ {{.
2. compiler/crates/hy-emit/src/rangebind.rs:
   • Because both inclusive and exclusive ForKind::Range loops now evaluate lo and hi in the enclosing scope before binding var, stmt_reads uniformly treats expr_reads(lo, target) || expr_reads(hi, target) as reading the outer target even when var == target.
3. Tests (compiler/crates/hy-emit/src/tests.rs & compiler/crates/hy-emit/tests/rangebind_match_soundness.rs):
   • Added exclusive_range_hoists_upper_bound_once_in_outer_scope verifying for n in 0..n lowers to hyHi0 := n followed by for n := 0; n < hyHi0; n++.
   • Updated rangebind_match_soundness.rs and golden emission assertions for the hoisted hyHi binding.

Verification

• cargo test -p hy-emit -p hy-check: 977 passed; 0 failed.