fix(compute/exprs): avoid empty field reference panic

[fallintoplace]

Rationale for this change

A field reference against an empty schema tries to include out.Type in the error before out has been set. This causes a nil pointer panic instead of returning ErrNoChildren.

What changes are included in this PR?

Return ErrNoChildren for an empty schema before formatting a field type. Preserve the existing error detail for nested references and add regression coverage.

Are these changes tested?

• go test ./arrow/compute/exprs

Are there any user-facing changes?

Invalid field references now return an error instead of panicking when the schema has no fields.

[zeroshade]

Correct and minimal.

out starts nil and is only assigned inside the StructFieldRef case, so on the first iteration with an empty field list the old code dereferenced a nil *arrow.Field to read out.Type — a panic in a function that otherwise returns errors. The guard is in the right spot.

Worth noting what this deliberately doesn't do: it would have been easy to hoist the emptiness check above the loop, but keeping it inside preserves the distinction between two different conditions — an empty schema at the top level, where there's no type to name, versus a nested reference into a field whose type has no children, where ErrNoChildren: <type> is genuinely useful. The nested path still reaches the original line with out non-nil, so that detail survives.

Both entry points are covered too, since GetRefFieldFromSchema funnels through here with schema.Fields() — an empty schema takes the same path, so the panic was reachable from the public API as well.

---

Note: this review was drafted with AI assistance by a maintainer and may contain mistakes. If anything here looks wrong, say so on the PR and I'll take another look.

[zeroshade]

Minor coverage gap: this pins the top-level empty case, but nothing pins the nested one. A future refactor that hoists the len(fields) == 0 check above the loop would silently drop the : <type> detail from nested references and this test would still pass.

A second case — a reference into a field whose type has no children, asserting the error still names the type — would guard the behaviour the fix went out of its way to preserve.