fix(hy-emit): a bare-error result is an error slot, not a value (#726)

[airdropia]

What

fn step() error + step()? passes hyb check, then the emitted Go fails
go build:

func run() error {
    hyT0, hyErr1 := step()        // step returns ONE value
    if hyErr1 != nil { ... }
}
// out.go: error: assignment mismatch: 2 variables but step returns 1 value

Why it happened

The fn-shape collector (hy-emit/src/lib.rs) recorded (f.results.len(), err, ...) — a trailing error result counted as a value AND as the error slot.
fn step() error has exactly one Go result, so the ? hoister's (1, true)
reading produced hyT0, hyErr1 := step(), which Go rejects. run()? in
main hit the same shape.

The fix

The value count now excludes the trailing error slot:

let value_count = if err { f.results.len() - 1 } else { f.results.len() };

The hoister's (0, true) special case (already present in stmt.rs —
hyErrN := call with no value temp) now applies to bare-error user functions
the same way it already applied to rust-crate methods like
serde_json.value.set(...).

Files: compiler/crates/hy-emit/src/lib.rs (one line + comment).

Tests

New emitter test bare_error_fn_under_try_emits_zero_value_temps pins the
emitted shape: hyErr0 := step() and no hyT0, hyErr1 := step().

No local Go toolchain: CI (parse + emit tests + go build of examples) is the
oracle.